mirror of
https://github.com/multica-ai/multica.git
synced 2026-07-28 14:09:22 +02:00
transcript-all
34 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Bohan Jiang
|
ffa8e16369 |
MUL-5228 fix(usage): bill Grok at xAI's reported cost, fix $0 resumed sessions (#5841)
* fix(agent): attribute Grok usage from the turn's own model id A resumed Grok session with no configured model recorded its entire spend under the model id "unknown", which matches no pricing row — so the task reported $0 cost instead of its real spend. grok.go only learned the model from the session handshake, and ACP's `session/load` carries no model id (only `session/new` does). When neither the agent nor MULTICA_GROK_MODEL pins a model, `daemon.go` legitimately passes an empty model, leaving nothing to attribute the usage to. Every Grok turn stamps `result._meta.modelId` with what it actually billed against. Parse it in the shared ACP result parser and use it as the fallback in grok.go. Other ACP backends are untouched — they keep whatever the handshake gave them. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * fix(metrics): price the Grok catalog in server-side cost metrics server/internal/metrics/pricing.go carried no Grok rows at all, so RecordLLMUsage took the unpriced branch for every Grok turn: llm_cost_usd reported zero Grok spend while the tokens accumulated in llm_unpriced_tokens. Internal cost monitoring simply could not see Grok. Add the six SKUs xAI publishes rates for, mirroring the frontend table in packages/views/runtimes/utils.ts. Aliases are anchored exact matches like the gpt-5.6 rows, so `grok-composer-*` (in the catalog, absent from the price sheet) stays unmapped instead of inheriting a guessed rate. Short-context tier on purpose: xAI bills a request at 2x once its prompt reaches 200K tokens, but a usage record aggregates every model call in a turn and cannot say which tier an individual request hit. A regression test re-derives the cost of a real grok 0.2.106 turn from the table and checks it against the costUsdTicks xAI returned for that turn. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * docs(changelog): scope the Grok cost claim to what was actually fixed The v0.4.9 entry promised "accurate cost" in all four languages, but the fix corrected catalog pricing and cached-input double-counting — it did not implement xAI's 2x long-context tier, so a turn whose requests reach 200K prompt tokens still under-reports by up to 50%. Say what was fixed instead. Also correct two stale claims in the pricing comment: the daemon tags usage rows with the runtime provider `grok`, not `xai` (the bare `grok-*` keys are what make them resolve), and record why thresholding the long-context tier on an aggregated row would be worse than not pricing it at all. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * feat(usage): carry the provider's own cost through to the usage record Cost has always been derived client-side as tokens x a static rate, which cannot express request-level pricing rules. xAI bills a Grok request at 2x once its prompt reaches 200K tokens, and a task_usage row aggregates every model call in a turn — so the stored token counts genuinely cannot say which tier any individual request hit. Thresholding on the aggregate would be worse than the status quo: it turns a bounded 50% under-estimate into an unbounded over-estimate for turns made of many short requests. Grok already reports what it charged, per turn, in `_meta.usage.costUsdTicks`. Parse it, carry it through agent -> daemon -> API, and store it on task_usage as a nullable BIGINT of 1e-10 USD ticks (integer, so sub-cent turns stay exact end to end). NULL means the provider reported no cost — every pre-existing row and every provider that doesn't return one. No backfill: there is no authoritative figure to recover for those, and inventing one is the guess this removes. A single hourly bucket can mix rows that carry a cost with rows that don't, so task_usage_hourly gains both halves: `cost_usd_ticks` sums the authoritative side, and `uncosted_*_tokens` carry exactly the tokens that still need a rate-table estimate. Consumers report authoritative + estimate(uncosted), which degrades to today's behaviour when nothing in the bucket is authoritative. The existing token columns keep covering every row, so token displays are untouched. The new columns are additive with defaults, so the unique key, the dirty-queue shape, and migration 102's triggers are unaffected. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * feat(usage): prefer the provider's own cost over the rate table With the authoritative figure now stored, both cost consumers use it: the usage dashboard (estimateCost / estimateCostBreakdown) and the server-side llm_cost_usd metric. Each reports `authoritative + estimate(uncosted tokens)`, so a row or bucket that mixes priced and unpriced sources stays whole. The static rate tables remain, but for Grok they are now a fallback — they still price usage recorded by a daemon too old to report cost, and every provider that reports none. Custom pricing overrides likewise apply only to the estimated half: they are a user's guess at a rate, and the authoritative half is not a guess. A model with no rate-table row but a provider-reported cost now also drops out of the "unmapped models" banner, since asking the user to supply a rate for it would invite overriding a real bill. llm_cost_usd is labelled by token_type and the provider reports one number per turn, so the charge is distributed across the buckets in the rate table's own proportions. Only the total is authoritative; the split stays an estimate, which is why this scales the existing buckets rather than inventing a label. estimateCostBreakdown does the same, keeping the stacked chart summing to the headline figure instead of silently under-drawing every Grok row. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * docs(changelog): say Grok cost now follows xAI's actual charge The earlier wording scoped the claim down to catalog pricing and cached input because the long-context tier was still unhandled. It is handled now — the cost comes from what xAI charged for the turn — so the entry can say so. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * fix(usage): keep the provider's cost when the model has no rate row Both cost consumers bailed out before reading the authoritative figure when the rate table had no row for the model. A `grok-composer-*` turn — in the Grok Build catalog, absent from xAI's price sheet — was therefore reported as $0 spend even though xAI told us exactly what it charged. Worse on the client: estimateCost returned the real cost while estimateCostBreakdown returned zeros, so the headline and the stacked chart disagreed on precisely the rows whose cost is exact — and the unmapped-models banner was (correctly) hidden, so nothing explained the discrepancy. Handle the charge before the rate lookup in both places. Without rates there is nothing to split a total by, so it lands whole in the `input` bucket, the same fallback distributeAuthoritativeCost already uses when it has no shape to scale. Tokens with no rate keep going to llm_unpriced_tokens: "unpriced" describes the rate table, not the money. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> * perf(usage): drop the historical rewrite from the cost-split migration Migration 213 rewrote every existing task_usage_hourly row to seed the uncosted counters. That is a full-table UPDATE inside a schema migration — lock time, WAL and bloat all scaling with table size — for rows this issue explicitly does not care about. Deleting the UPDATE alone would have zeroed historical cost: with `NOT NULL DEFAULT 0`, an untouched row asserts "nothing here needs estimating", so every pre-split bucket would report $0 until the rollup happened to touch it. Make the uncosted columns nullable with no default instead. NULL means "never recomputed since the split existed", readers COALESCE it to the row's own token total ("estimate all of it"), and the pre-split behaviour is preserved exactly — with nothing to seed, so no rewrite. A bare ADD COLUMN is metadata-only, so this is now fast DDL. Rows heal into the split naturally as the rollup recomputes their buckets. Verified on a fresh database: a legacy-shaped row reads back as its full tokens to estimate, and a group mixing legacy and post-split buckets sums to the authoritative cost plus both rows' estimable tokens. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: Bohan-J <bohan@devv.ai> Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Bohan Jiang
|
ecbdbda09e |
fix(agent): stop double-counting Grok cached input tokens (#5838)
Grok Build reports cachedReadTokens inside inputTokens (totalTokens == input + output on a real 0.2.106 turn, and that turn's costUsdTicks matches xAI's rates only when the cached prefix is billed once). The shared ACP parser persisted both counters raw, so the usage dashboard charged the cached prefix at the full input rate *and* the cache-read rate — ~4x the real spend on a cache-heavy turn. Re-bucket cached reads out of input when totalTokens proves the overlap, the same normalization codex.go already applies. Backends that report mutually-exclusive buckets or omit totalTokens are untouched. Co-authored-by: J <agent@multica.ai> Co-authored-by: multica-agent <github@multica.ai> |
||
|
chenow9
|
d804dedcc6 |
MUL-5137: fix(agent): parse Grok ACP token usage from session/prompt _meta (#5748)
* fix(agent): parse Grok ACP token usage from session/prompt _meta Grok Build places per-turn metering under result._meta (and _meta.usage), not the top-level ACP usage field. Multica's shared parser only read result.usage, so Grok tasks recorded empty token usage and cost dashboards stayed at zero. Fall back to _meta.usage (then flat _meta counters) when top-level usage is absent. Prefer standard top-level usage when both are present. Update the Grok fake ACP fixture and add regression tests against the live 0.2.x payload shape. * test(agent): cover zero ACP usage meta fallback Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: Eve <eve@multica-ai.local> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Bohan Jiang
|
a5a42846e6 |
fix(daemon): retry with a fresh session only when the resume was actually rejected (MUL-4966) (#5715)
* fix(daemon): gate fresh-session retry on tools executed, not session id (MUL-4966) Switching provider accounts leaves the stored session id pointing at a conversation the new account does not own. The daemon still passes it to --resume, the provider rejects it, and the task dies before doing any work. The existing fresh-session fallback was supposed to catch this but was gated on `result.SessionID == ""`, which is not a lifecycle fact: - Too narrow: a backend that echoes the requested id back when it rejects a resume keeps SessionID non-empty, so the fallback never fired — the reported bug. - Too broad: a provider 401 before the first stream message also leaves SessionID empty, so an unrecoverable auth failure burned a second full run. Gate on `tools == 0` instead. That states the property that actually makes a retry safe — the agent executed no tool, so it mutated nothing, so re-running cannot double-post a comment (comment creation has no idempotency key and a duplicate re-fires its @mention triggers), reopen a PR, or re-plan on top of its own half-finished work in the reused workdir. Auth failures are excluded, mirroring retryableReasons in service/task.go. The predicate is extracted to shouldRetryWithFreshSession so the tests exercise production logic; both existing fallback tests re-implemented the condition inline and would not have caught a regression in it. Co-authored-by: multica-agent <github@multica.ai> * fix(daemon): gate fresh-session retry on a positive resume-rejected signal (MUL-4966) Review of the previous commit was right: `tools == 0` plus "not an auth error" answers whether re-running is *safe*, not whether a new session can *fix* the failure. Those are orthogonal, and answering the second by exclusion inverts the burden of proof — the failures a fresh session cures are a small enumerable set, while the ones it cannot are open-ended. Concretely, the previous predicate fresh-retried on provider_network, 429/529, quota, 5xx and unclassified startup failures. provider_network is the sharpest conflict: internal/service/task.go marks it resume-safe (MUL-4910) specifically so the platform retry inherits the session and continues the truncated conversation. Resetting the session first made that contract unsatisfiable, silently discarding conversation context on a transient blip — and rate limits got an immediate no-backoff re-run. Replace the inference with positive evidence: agent.Result gains an explicit ResumeRejected field, set only when a backend has proof the resume itself was refused. claude/codebuddy/qwen derive it from resumeWasRejected, which promotes the predicate resolveSessionID was already computing and encoding as the side effect of blanking SessionID — using an empty string to carry that meaning is what made the original bug possible. SessionID keeps being dropped for a rejected resume (a dead pointer must not be persisted), but it is no longer the signal the daemon reads to decide *why* a run failed. The six ACP backends that recover from "session not found" set the flag at the same points they already clear the id, so their existing recovery is not caught by the narrower gate. codex needs nothing: thread/resume already falls back to thread/start in-process, and deliberately does not on transport errors. Matching now includes the account-switch guardrail reported in #5704 (Claude Code 2.1.207, zh-CN): "400 此 session 已绑定另外的ai账号，请执行 /new 开启新 session". The en-US wording of the same guardrail has not been captured yet, so those variants are marked inferred in the source; a miss degrades to a terminal failure carrying the provider's raw text rather than a mis-routed run. Tests: backend-level fixtures drive ResumeRejected from real stream-json for both the account-binding 400 and a network drop, and the predicate now covers network/rate-limit/quota/5xx/auth/unclassified as explicit non-retries. Co-authored-by: multica-agent <github@multica.ai> * fix(daemon): restore fresh-session recovery for backends with no rejection signal (MUL-4966) Final review caught qwen regressing: its verified rejection string ("No saved session found with ID ...", already captured in testdata/qwen-code-0.20.0-resume-not-found.stderr.txt) was not in the phrase list, and qwen reports no session id on that path, so the new inclusion gate turned a working auto-recovery into a terminal failure. Auditing the other 17 resume-capable backends showed qwen was not alone. antigravity, copilot, cursor, deveco and opencode all recovered from a refused resume purely by reporting an empty SessionID, and none of them has any rejection detection to convert into ResumeRejected — copilot's own comment documents the hole (session.error before session.start), and antigravity's helper returns "" when "the CLI exited before dispatching". Making ResumeRejected the sole gate silently removed recovery from all five. Fixing that by guessing rejection phrases for five more CLIs is the wrong trade: no real output has been captured for any of them, and a false positive discards a recoverable session pointer. So the gate is now two tiers. Positive evidence (ResumeRejected) decides on its own where a backend can produce it. Where none is available, an empty SessionID still gates the retry — it proves no session was established, which is exactly what the pre-change behaviour relied on — minus the classes a fresh session provably cannot cure (network, rate limit, quota, provider 5xx, auth). That keeps the resume-safe contract in internal/service/task.go intact while restoring what these five backends had. Also renames claudeResumeRejectedPhrases to resumeRejectedPhrases: it is matched by claude, codebuddy and qwen, so a qwen-only string living under a claude-prefixed name would be actively misleading. Tests: qwen's existing missing-resume fixture now asserts ResumeRejected (verified failing without the phrase), and the predicate covers the no-signal tiers — retry when nothing was established, no retry once a session exists or the failure classifies as uncurable. Co-authored-by: multica-agent <github@multica.ai> * fix(daemon): scope the no-signal fallback to backends that cannot detect rejections (MUL-4966) Final review caught the compatibility path applying to every backend, not just the five it was justified for. shouldRetryWithFreshSession only saw (Result, priorSessionID, tools), so a false ResumeRejected could not be told apart from a backend that has no way to answer — and claude/codebuddy/qwen/ACP startup failures with no session id still fell through to the exclusion branch. That contradicted both the stated intent and the function's own doc comment ("where a backend can produce it, it is the whole answer"). Make the capability explicit. agent.ResumeRejectionUndetectable names the five backends that scrape SessionID out of stream output and have no rejection detection at all; the daemon takes provider and consults it, so a capable backend reporting false is now taken at its word. Membership is opt-in, so a new backend fails closed instead of silently inheriting a guess-based retry. Also completes the exclusion set: missing config, unavailable model, missing executable, unsupported runtime version and (defensively) agent timeout all have defined non-session remedies and were reaching `default: true`. What is left through stays narrow — unknown, process failure, unparseable output, context overflow — because a real rejection from these five most likely surfaces as a non-zero exit or unparseable output, none of them reporting one explicitly. Tests: one identical result asserted across all five undetectable backends (retries), twelve capable ones (no retry), and an unregistered provider (fails closed), plus table cases for each newly excluded reason. Classifier inputs were verified to map to the intended reasons rather than passing by accident. Also updates the ResumeRejected doc comment, which still said the daemon gates on it alone. Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: Bohan-J <bohan@devv.ai> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Bohan Jiang
|
e2f4f28462 |
fix(agent): skip redundant Hermes set_model when session already on the requested model (#5690)
When agent.model equals the model the Hermes ACP session already reports as current, Multica was still replaying session/set_model. Hermes' set_model re-runs provider auto-detection on the model id, which for a provider:model id whose parsed provider matches the session's current provider (e.g. a named custom endpoint exposed as "custom") can mis-route to a different provider — custom:deepseek-v4-pro resolves into the OpenRouter catalog and the turn fails with an auth error. Only the retry via session resume, where the corrupted provider no longer matches the custom: prefix, succeeds. Capture the runtime's reported currentModelId from session/new and session/resume and skip the redundant switch when it already equals the requested model. An empty/unparsable current model falls through and still sends set_model, preserving prior behaviour. The explicit provider:model id is never rewritten. Upstream: NousResearch/hermes-agent#59089. MUL-5029 Co-authored-by: Bohan-J <bohan@devv.ai> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Multica Eve
|
f51828f568 |
MUL-4936: Fix empty replies in resumed Codex and Hermes chats (#5675)
* fix(agent): isolate resumed Codex turns (MUL-4936) Co-authored-by: multica-agent <github@multica.ai> * fix(agent): drain late Hermes replies (MUL-4936) Co-authored-by: multica-agent <github@multica.ai> * fix(agent): harden resumed chat cleanup (MUL-4936) Co-authored-by: multica-agent <github@multica.ai> * fix(agent): filter Codex subagents before turn gate (MUL-4936) Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: Eve <eve@multica-ai.local> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Multica Eve
|
f0d7a444ac |
MUL-4860 fix(acp/kiro): GPT-5.6 Sol completion — object rawOutput + positive-proof guard (follow-up to #5511) (#5515)
* fix(kiro): require positive proof for -32603 completion rescue (MUL-4860, #5509) Addresses the #5511 review (Elon): Must-fix 1 — a missing result is not proof of success. A mid-command crash / internal session failure produces the same 'tool use, no result, -32603' shape as a genuine completion, so the previous 'saw the use but never saw a result' rescue could mark an unfinished task completed. The guard now rescues only on POSITIVE proof: a terminal ToolResult with status=='completed' for a finishing tool (goal_complete / comment-add). Must-fix 2 — the previous sticky booleans only ever wrote true, so a later failed delivery could not override an earlier success, and call ordering was ignored. Completion state is now the status of the MOST RECENT finishing-tool result (keyed per CallID): 'progress completed → final failed → -32603' stays failed; 'first failed → retry completed → -32603' is a real completion. Comment-add is still recognized by its command payload regardless of the tool's display title (the #5509 core ask), so a completed comment-add carrying a non-terminal title is preserved through the close handshake. Tests: result-less running tool stays failed (both goal_complete and comment-add), completed non-terminal-title comment-add is preserved, failed-final-overrides-earlier-success, and completed-retry-after-failure. Co-authored-by: multica-agent <github@multica.ai> * fix(acp): don't drop completed tool_call_update when rawOutput is an object (MUL-4860, #5509) Captured a live kiro-cli 2.12.3 + gpt-5.6-sol ACP trace and found the true root cause: the completed tool_call_update sends rawOutput as an OBJECT ({"items":[{"Json":{...}}]}), but handleToolCallUpdate typed rawOutput/output as Go strings. json.Unmarshal then failed with 'cannot unmarshal object into Go struct field .rawOutput of type string' and the handler returned early — silently dropping the ENTIRE update, status:"completed" included. So no completion signal ever reached the kiro completion-preservation guard, and the durable-but-closed task was marked failed. This also explains the issue's 'shell tool recorded as running': the tool_call title is literally 'Running: <cmd>' (kind execute), which normalizes to 'running', not 'terminal'. Fix: type rawOutput/output as json.RawMessage and render them via a new acpRawText helper that accepts both a JSON string and a structured value. This is a shared ACP-layer fix (hermes/kimi/kiro). Combined with the earlier payload-based comment-add recognition and the positive-proof completion guard, the completed result now flows through and the -32603 close handshake correctly preserves completion. Adds TestKiroBackendPreservesCompletionOnRealGPT56SolFrames (the exact captured wire shape end-to-end) and TestACPRawText. Verified the end-to-end test fails (status=failed) when rawOutput is typed as string and passes after the fix. Co-authored-by: multica-agent <github@multica.ai> * fix(kiro): remove duplicate stale-session branch that faked completion (MUL-4860, #5509) Addresses Elon's 3rd-round review of #5515. The -32603 guard had two consecutive, identically-conditioned else-if branches for a stale resumed session (opts.ResumeSessionID != "" && isACPSessionNotFound(err)). The first wrongly forced finalStatus="completed" and kept the stale SessionID; the second — the correct handler that clears SessionID so the daemon's fresh-session retry fires — was unreachable. So a recovery whose resumed session was gone at session/prompt time was reported as a fake success AND skipped the retry. Remove the bogus first branch, keeping the positive-proof completion branch and the original stale-session (SessionID="") branch. Adds TestKiroBackendClearsSessionIDWhenPromptSessionNotFound (the prompt path; existing coverage only exercised session/set_model). Verified it returns completed+ses_stale on the pre-fix code and failed+empty SessionID after. Also gofmt on the touched files. Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: Eve <eve@multica-ai.local> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Multica Eve
|
66316c2614 |
fix(grok): harden ACP authentication and capabilities (#5440)
Co-authored-by: Eve <eve@multica-ai.local> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Sanjay Ramadugu
|
06d79c1750 |
feat(agent): add Grok Build CLI as an ACP runtime (#5285)
Add xAI Grok Build (`grok`) as a first-class Multica runtime over ACP (`grok --no-auto-update agent --always-approve stdio`), reusing hermesClient like traecli and kimi. Includes daemon discovery, protocol_family migration 174, model discovery, MCP passthrough, thinking effort, frontend branding, and product docs. Follows xAI's documented headless ACP flow: after `initialize`, read the advertised `authMethods` and send `authenticate` (preferring xai.api_key when XAI_API_KEY is set, else the cached login token) before any session operation — a real, logged-in CLI rejects session/new and session/load without it. Model discovery performs the same handshake so it returns the live catalog instead of the static fallback. `--no-auto-update` is passed as a global flag (and kept daemon-owned in the blocked custom-arg set) so a background update check can't stall an unattended ACP task. Thinking effort uses the current `--effort` flag, and the minimum grok version is 0.2.89 (ACP + authenticate + session/load + session/set_model + MCP + --effort). Closes #2895 |
||
|
Rusty Raven
|
ef9b334408 |
MUL-4398: fix Hermes bound-skill discovery with per-task overlay (#5308)
* fix(execenv): overlay per-task HERMES_HOME so Hermes discovers bound skills Hermes has no workspace-relative skill discovery — it scans <HERMES_HOME>/skills first, then skills.external_dirs from config.yaml (verified against the bundled agent/skill_utils.py). The daemon wrote assigned skills to the generic .agent_context/skills/ fallback, which Hermes never reads, so they silently never took effect (#5242). When (and only when) an agent has skills bound, redirect HERMES_HOME to a minimal per-task compatibility overlay; a skill-less Hermes task keeps its real home and original behavior: - mirror every top-level entry of the shared home via symlink except the overlay-owned ones (denylist), reconciling entries deleted from the shared home; - derive a task-local config.yaml whose skills.external_dirs references the shared skills dir plus the user's existing external_dirs, expanded against the sanitized effective child env (unknown vars preserved, blocklisted keys resolved to the process value) and normalized to absolute paths; - write only the bound skills into the task-local skills/ dir (home skills scanned first, so they win); global skills are referenced, not copied; - keep memories/ overlay-owned (fresh per-task dir) AND disable the external memory.provider, so neither on-disk memory nor a shared backend crosses tasks; - keep active_profile/profiles out of the overlay so Hermes can't follow a sticky profile and redirect past it at startup. Profile handling mirrors hermes_cli.profiles: the daemon reads -p/--profile with agent.HermesProfileFromArgs and seeds the overlay from that profile's home via ResolveHermesSourceHome (default/invalid -> base, valid name -> <base>/profiles/ <name>, validated; a missing named profile fails closed). The profile flags are stripped from the acp argv ONLY when the overlay is active (hermesLaunchArgs), so a skill-less task's profile passes through unchanged. HERMES_HOME is no longer custom_env-blocklisted: no skills -> user value passes through; skills -> overlay overrides after layering. Fail closed — Prepare errors, Reuse returns nil. Task home 0700, derived config 0600 via atomic replace. Platform-native default home (%LOCALAPPDATA%\hermes, incl. the LOCALAPPDATA-missing fallback, on Windows). Tests span execenv/daemon/agent: no-skill no-op, child-env layering + env sanitization, profile parse/unquote + conditional strip + final args/env per scenario, custom/profile/default/invalid/missing/Windows source home, sticky- profile not mirrored, memory dir isolation + external provider disable, mirror reconciliation, external_dirs rebasing + sanitized/unknown-var expansion, local-precedence slug, perms, fail-closed, resume teardown. Docs (en + ja/ko/zh). Fixes #5242 * fix(hermes): make profile selection one resolver contract matching Hermes Round 5 review: the profile chain approximated Hermes' semantics in three separate places (argv parsing, source-home selection, arg filtering), so it diverged from native Hermes in several merge-blocking cases. Collapse it into one authoritative resolution: - agent.ParseHermesProfileArgs replaces HermesProfileFromArgs/ FilterHermesProfileArgs. It reproduces _apply_profile_override step 1/1b (first occurrence, value-flag skipping, `--` and `mcp add --args` boundaries, space-form profile-id guard) and returns the exact argv occurrence to consume; StripHermesProfileArgs removes only that occurrence. - execenv.ResolveHermesProfile replaces ResolveHermesSourceHome. It derives the Hermes root exactly like get_default_hermes_root (an already-profile-scoped HERMES_HOME roots at its grandparent), selects an explicit profile first, otherwise trusts a profile-scoped home (step 1.5) and only then the sticky <root>/active_profile (step 2), and validates via normalize/validate_profile_name (reserved hermes/test/tmp/root/sudo and empty inline `--profile=` are hard errors). Profiles always resolve under the root, so `-p default` re-roots and `-p <sibling>` is a sibling, never nested. - The daemon runs one parse + resolve, fails the task closed on a reserved/ invalid selection (matching Hermes' sys.exit(1)), and exports the selected source home as the effective env's HERMES_HOME so ${HERMES_HOME} in a profile's skills.external_dirs expands against the selected profile home (as native Hermes does before loading config.yaml), not the root or the overlay. Regressions added: root + sticky named profile selection; already-profile-scoped home with no flag; that home with -p default and -p <sibling>; reserved and empty inline profile values; and a selected profile whose external_dirs contains ${HERMES_HOME}. * fix(hermes): overlay-owned derived .env + symlink-resolved root Round 6 review, two remaining overlay-bypass paths: 1. A source `.env` could redirect HERMES_HOME after profile resolution. Hermes runs `_apply_profile_override()` then `load_hermes_dotenv()`, which loads `<HERMES_HOME>/.env` with override=True — so a mirrored source `.env` carrying an out-of-band `HERMES_HOME=` overwrote the overlay's home, repointing skill discovery and memory back at the source. `.env` is now overlay-owned and DERIVED (writeDerivedHermesEnv): it preserves the source's credentials/settings but strips any `HERMES_HOME` assignment and pins `HERMES_HOME` to the overlay last (single-quoted, literal), written 0600 via atomic replace. It is written even when the source has none, so Hermes' project-`.env` fallback (override=True only when no user `.env` loaded) can't relocate the home either. 2. Root derivation was lexical-only, diverging from `get_default_hermes_root`, which compares `env_path.resolve()` with `native_home.resolve()`. A HERMES_HOME symlinked into `<native>/profiles/<x>` was treated as its own root, so `-p default`/`-p <sibling>` resolved wrong. `hermesRootFromHomeFor` now resolves symlinks (Path.resolve(strict=False)-style best effort) for the containment decision while keeping the returned root unresolved, matching Hermes. Regressions: source `.env` with HERMES_HOME replayed through the override=True dotenv order (bound skill + task memory stay on the overlay; creds preserved); minimal overlay `.env` created when the source has none; and a symlinked profile home resolving `-p default`/`-p <sibling>` to the native root. |
||
|
Bohan Jiang
|
7a1a2a9de4 |
fix(agent): select an offered ACP permission option so Hermes writes aren't denied (MUL-4441) (#5351)
Fixes #5300. The daemon hardcoded optionId="approve_for_session" when auto-approving Hermes' session/request_permission, but Hermes' ACP edit-approval offers only ["allow_once","deny"] and rejects anything else, silently blocking every file write. handleAgentRequest now selects an option the agent actually offered — a safe session/single-use grant, else an offered reject_once to deny just that action, else a JSON-RPC error — never a permanent allow_always or a whole-turn cancelled. Includes regression + branch-coverage tests. |
||
|
ZeroIce
|
1e85eb0aac |
Fix Kiro ACP usage accounting (#4867)
Co-authored-by: multica-agent <github@multica.ai> |
||
|
Multica Eve
|
b71d9d0ab9 |
MUL-3674: Preserve Kiro goal completion on close error (#4560)
* fix(daemon): preserve Kiro goal completion on close error Co-authored-by: multica-agent <github@multica.ai> * fix(daemon): require completed Kiro goal marker Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: Eve <eve@multica-ai.local> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Truffle
|
6acca84c28 |
fix(agent): clear stale session id when a resumed ACP session is gone [MUL-3216] (#4015)
* fix(agent): clear stale session id when a resumed ACP session is gone When an agent's stored ACP session no longer exists on the runtime side, session/resume still succeeds — hermes echoes the requested sessionId back — so the failure only surfaces when session/prompt returns JSON-RPC -32603 "Session not found". The backend then reported Status=failed with the stale SessionID still set, which kept the daemon's resume-failure fallback (gated on SessionID == "") from ever firing. The failed task never updates the stored session, so every future mention on the same (agent, issue) dispatched against the same dead id, forever (#4010). handleResponse now returns a structured acpRPCError instead of a flat string (rendered text unchanged), and the hermes/kimi/kiro prompt-error paths clear the session id when the error is session-not-found class on a resumed session. The daemon's existing retry then re-executes with a fresh session and stores the replacement id, healing the mapping. * fix(agent): clear stale session id when set_model hits a dead resumed session With a model override, session/set_model runs before session/prompt, so a resumed session that is gone on the agent side surfaces there instead of at the prompt — and the error branch returned the stale SessionID, so the daemon's fresh-session retry (gated on SessionID == "") never fired. Apply the same clear-the-id fix in the set_model error branch of all three backends. Also relax isACPSessionNotFound to accept -32602: kimi-cli raises RequestError.invalid_params({"session_id": "Session not found"}) for every unknown-session path (src/kimi_cli/acp/server.py), so pinning -32603 made the fix dead code for kimi. The wording gate keeps unrelated invalid_params errors (e.g. "model not available") on the preserve-the-id path. Regression tests for all three backends: resumed session + model override + set_model failing with each runtime's observed session-not-found shape must yield status=failed with an empty SessionID. |
||
|
Bohan Jiang
|
3708fb0f07 |
fix(daemon): inactivity-based agent run timeout, no wall-clock guillotine (MUL-3064)
Active long-running sessions are no longer killed by a fixed wall-clock deadline. Liveness is delegated to the idle watchdog (MULTICA_AGENT_IDLE_WATCHDOG, default 30m) with a larger in-flight-tool budget (MULTICA_AGENT_TOOL_WATCHDOG, default 2h). MULTICA_AGENT_TIMEOUT is an opt-in absolute cap (default 0 = no cap). The server-side 2.5h sweeper is unchanged as a coarse backstop. Fixes #3745. |
||
|
Bohan Jiang
|
09f9c7e2ce |
MUL-2764 feat(agent): wire mcp_config through ACP runtimes (Hermes / Kimi / Kiro) (#3439)
* MUL-2764 feat(agent): wire mcp_config through ACP runtimes (Hermes / Kimi / Kiro) The MCP config Tab (#3419) already lets admins save mcp_config on an agent, and the daemon plumbs it through to `agent.ExecOptions.McpConfig` for every runtime. Claude and Codex consume it; the three ACP runtimes (Hermes / Kimi / Kiro) ignored the field and hardcoded an empty `mcpServers: []` in their `session/new` requests. Add `buildACPMcpServers` to translate the Claude-style `{"mcpServers": {"<name>": {...}}}` object-of-objects into the array shape ACP requires (`[{name, command, args, env: [{name,value}, ...]}, ...]` for stdio; `[{type, name, url, headers: [...]}, ...]` for http/sse), then pass the translated array on `session/new` (all three) and `session/load` (kiro resume). Malformed JSON fails the launch closed — same contract Codex's `renderCodexMcpServersBlock` uses — so users see a real error instead of silently running with no MCP servers. Individual unclassifiable entries (no command, no url) are skipped with a warning so one bad row can't take MCP down for the rest of the agent. Co-authored-by: multica-agent <github@multica.ai> * MUL-2764 fix(agent): wire mcp_config through ACP resume + gate http/sse on capability Addresses the two blockers Elon raised on #3439: 1. session/resume now carries mcpServers for Hermes and Kimi (Kiro's session/load already did). Per the ACP Session Setup spec the resume path re-attaches MCP servers, and without this a resumed task lost access to MCP tools that a fresh task on the same agent would have had. Pinned with new TestHermesResumeIncludesMcpServers and TestKimiResumeIncludesMcpServers integration tests that inspect the recorded wire request. 2. Added extractACPMcpCapabilities + filterACPMcpServersByCapability so http/sse MCP entries get dropped (with a daemon-log warning naming the entry) when the runtime's initialize response doesn't advertise mcpCapabilities.http / .sse. Sending those entries to a stdio-only runtime is a spec violation and reliably tanks session/new; now they get filtered and the rest of the session still starts. Stdio entries pass through unconditionally. Both backends wire the filter in right after initialize so session/new and session/resume see the same filtered list. Also added TestKiroLoadIncludesMcpServersFromConfig — Elon flagged that no test pinned "non-empty mcp_config actually reaches the wire" for Kimi/Kiro, so the wire assertions go in for all three runtimes. Co-authored-by: multica-agent <github@multica.ai> --------- Co-authored-by: J <j@multica.ai> Co-authored-by: multica-agent <github@multica.ai> |
||
|
Multica Eve
|
26ff52385b |
fix: attribute Hermes usage to current model (MUL-2696)
Fix Hermes ACP usage attribution to current model when agent.model is unset. Also preserves cache-read token accounting and makes ACP model-list parsing more tolerant of snake_case payloads and Unknown display names. |
||
|
fr00st
|
cc9fbd3db0 |
Fix stale Done replies on comment follow-ups (#2495)
* fix: avoid stale done replies on comment follow-ups * fix: avoid inlining runtime brief for Hermes ACP * fix: address comment follow-up review feedback |
||
|
Bohan Jiang
|
b73a301bf9 |
fix(agent): drain stderr before deciding ACP failure promotion (#2333)
`hermes`, `kimi`, and `kiro` all wired stderr through
`cmd.Stderr = io.MultiWriter(logWriter, providerErrSniffer)`.
The OS-pipe → MultiWriter copy goroutine that exec spawns for
that form is only joined by `cmd.Wait()`, which the lifecycle
goroutine fires in deferred cleanup — *after*
`promoteACPResultOnProviderError` already consulted the sniffer.
When stopReason=end_turn (success) raced ahead of the stderr
drain, the sniffer's `lines` slice was empty, the helper fell
through to the synthetic agent-text fallback ("hermes provider
error: API call failed after 3 retries"), and the actionable
upstream signal (HTTP 429 / usage limit) was lost.
This was visible as a flaky
`TestHermesBackendPromotesProviderErrorWithNonEmptyOutput` in CI
under high parallelism — a real prod bug, not a test issue: live
runs hit the same race when an upstream LLM returns 429 and
hermes' synthetic agent turn beats the stderr drain to the
parent.
Replace the MultiWriter wiring with `cmd.StderrPipe()` + an
explicit copier goroutine that signals on `stderrDone`. The
lifecycle goroutine already awaits `<-readerDone` for stdout;
add `<-stderrDone` next to it before `promoteACPResultOnProviderError`
runs. The deferred `cmd.Wait()` ordering is unchanged — it just
becomes a cheap reap by the time it fires.
Verified: `go test ./pkg/agent/ -run "TestHermes|TestKimi|TestKiro"
-count=10 -race`, then full package `-count=3 -race`, all green.
Co-authored-by: multica-agent <github@multica.ai>
|
||
|
LinYushen
|
f70105fb12 |
fix(agent): include JSON-RPC error data field in ACP error messages (#2327)
ACP backends (Kiro, Hermes, Kimi) put the actionable reason for code=-32603 'Internal error' in the JSON-RPC `data` field, e.g. "No session found with id". The wrapped Go error only carried `code` and `message`, leaving operators staring at a bare "kiro session/prompt failed: session/prompt: Internal error (code=-32603)" with no way to tell apart session expiry, model unavailability, lost auth, or quota. Parse `data` too. Strings render unquoted; objects/arrays render as raw JSON; null/missing keeps the previous format unchanged. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> |
||
|
Bohan Jiang
|
c57546159d |
fix(daemon): mark provider 429 / out-of-credit agent runs as failed, not completed (#2323)
* fix(daemon): mark provider 429 / out-of-credit runs as failed, not completed
Two bugs combined to silently report failed agent runs as
"Completed" in the UI when the upstream LLM returned a 4xx (e.g.
HTTP 429 rate-limit / no credit on the account).
1. ACP backends (hermes, kimi, kiro) only promoted the run status to
"failed" when their stderr sniffer fired AND the agent output
buffer was empty. But hermes injects a synthetic agent text turn
("API call failed after 3 retries: HTTP 429...") on retry
exhaustion, so the buffer was never empty in the rate-limit
case and the promotion never ran. Drop the empty-output
precondition: the sniffer's regex (HTTP-status markers, named
error types) is specific enough to trust on its own.
2. The daemon's task-result switch only routed "blocked" through
FailTask; every other status — including "cancelled", and any
future status we forget to enumerate — fell through to
CompleteTask. Invert it so only an explicit "completed" status
reports success, and extract the switch into reportTaskResult
for direct testing. Cancelled now defaults to failure_reason
"cancelled" instead of being silently completed.
Closes GitHub multica#1952.
Co-authored-by: multica-agent <github@multica.ai>
* fix(agent): only promote ACP run to failed on terminal provider error
Address GPT-Boy's review on the multica#1952 fix. The previous
promotion rule ("any sniffer line → fail") was too broad: the
existing sniffer also captures transient per-attempt warnings
("API call failed (attempt 1/3): RateLimitError [HTTP 429]"), and
those lines stay in the buffer for the rest of the run. A retry
sequence whose first attempt blipped but whose third attempt
succeeded would have been wrongly reported as failed.
Tighten the criteria with two additional signals, both defined on
the existing acpProviderErrorSniffer / output buffer:
- acpTerminalErrorRe — sticky `terminal` flag set when stderr shows
an exhausted/non-retryable marker (❌, [ERROR], "after N retries",
Non-retryable, BadRequestError, AuthenticationError). Per-attempt
warnings deliberately don't match.
- acpAgentOutputTerminalRe — matches the synthetic "API call failed
after N retries..." turn that hermes-style adapters inject into
the agent text stream when they give up; this catches multica#1952
even if hermes' stderr only logged transient attempts.
Promotion logic becomes a shared helper, promoteACPResultOnProviderError,
called from hermes / kimi / kiro. Promotes when (a) terminalMessage
is non-empty, (b) output contains the synthetic give-up turn, or
(c) output is empty and the sniffer captured anything at all
(preserves the original empty-output safety net for transient-only
sequences with no real result to fall back on).
Tests:
- TestHermesProviderErrorSnifferTerminalVsTransient — transient
attempt 1/3 alone returns terminalMessage="" but message!="";
a follow-on terminal marker flips terminal on.
- TestHermesProviderErrorSnifferTerminalNonRetryable — confirms
BadRequest / Authentication / Non-retryable / ❌ / [ERROR] are
classified terminal even on the very first attempt.
- TestHermesBackendDoesNotPromoteOnTransientRetry — fake hermes
emits attempt 1/3 to stderr then a normal agent text turn and
end_turn; resulting Status must stay "completed".
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
|
||
|
prellr
|
ee10c508fb |
fix(daemon): trust the agent's session id from session/resume across ACP backends (#2070)
When the local state.db of an ACP backend (hermes, kimi, kiro) is wiped — crash, config change, manual kill, container reset — the backend's session/resume (or session/load, in kiro's case) silently creates a brand-new session rather than failing, and returns the new id in the response. Today the daemon ignores the response and stamps sessionID = opts.ResumeSessionID across all three backends, so every subsequent session/prompt is addressed to a session id the backend has no record of. The task fails with JSON-RPC -32603 (Internal error) on the very first turn, with no operator-visible signal that the problem is a session-id mismatch one layer down. The behavior is invisible: agent shows "started", then "failed" with a generic Internal error. Reproducing in production took repeated runs because nothing in the logs pointed at the silent reset. Fix: route all three ACP backends through a small `resolveResumedSessionID` helper that: - prefers the id the backend returned in its response (the canonical id; the one the backend will accept on the next call) - falls back to the requested id when the response is malformed, empty, or omits sessionId — defensive fallback so older / non- conforming backends (notably kiro's current session/load shape) behave identically to today - signals (via a bool) when the id changed, so the caller logs a Warn with `backend=<hermes|kimi|kiro>` and operators can grep for silent state resets to correlate them with task failures Why this is at the backend layer rather than the daemon's existing session-resume fallback: server/internal/daemon/daemon.go:1554-1566 already retries with a fresh session when resume fails, but it gates on `result.Status == "failed" && result.SessionID == ""`. The backend WILL hand back a result.SessionID — just the new one it silently committed to — so the daemon-level fallback never fires for this failure mode. The helper is also what session/new already uses (extractACPSessionID, documented in code as "Shared by all ACP backends"). session/new extracts the canonical id from the response; session/resume just didn't, until now. Coverage: - hermes.go: confirmed bug, root cause of -32603 in production - kimi.go: same code shape, same protocol method, same response schema as hermes (per extractACPSessionID's comment) — same bug - kiro.go: same code shape, different method (session/load). Current observed response doesn't include sessionId, so the defensive fallback means today's behavior is preserved. Routing through the same helper means a future kiro release that DOES return a sessionId on silent reset works the same way as hermes/kimi without another diff. Tests (server/pkg/agent/hermes_test.go — helper covers all three backends, no per-backend duplication): - TestResolveResumedSessionIDMatching — backend confirms requested id - TestResolveResumedSessionIDDifferent — backend returned a new id; caller is told to switch - TestResolveResumedSessionIDEmptyResponse — older / malformed body; defensive fallback to requested id (covers kiro's current shape) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Bohan Jiang
|
170fa2102b |
fix(agent/hermes): wire streamingCurrentTurn gate to drop history replay (#2024)
Hermes ACP can flush queued session updates from the previous turn before the current turn actually starts — both as session/resume history replay and as chunks queued before our session/prompt response streams. Without a gate those updates were appended to output and re-emitted to the UI, so the previous answer appeared duplicated next to the new one. Closes #1997. PR #1789 added the acceptNotification hook field to hermesClient and the call site in handleNotification, but never assigned it for Hermes, so the guard short-circuited and every notification was processed. This change mirrors the working Kiro pattern (kiro.go:87/97/240): - declare a streamingCurrentTurn atomic.Bool in the backend. - assign acceptNotification, onMessage, onPromptDone gates that all return early when the flag is false. - flip the flag to true immediately before c.request("session/prompt"). Adds TestHermesClientAcceptNotificationGate as a regression test that exercises the gate directly on hermesClient. Verified with `go test ./pkg/agent`. Co-authored-by: multica-agent <github@multica.ai> |
||
|
LinYushen
|
03f3180b8f |
fix(agent): ignore Kiro session/load history replay (#1789)
Ignore Kiro ACP session/load history replay before the active prompt starts; keep task messages, usage, and tool state scoped to the current Kiro turn. Verified with go test ./pkg/agent -run TestKiro, go test ./pkg/agent, and git diff --check origin/main...HEAD. |
||
|
LinYushen
|
c366cf2ba1 |
feat(agent): add Kiro CLI ACP runtime (#1780)
* feat(agent): add kiro cli acp runtime * fix(agent): align kiro acp prompt and notifications * chore(agent): clarify kiro acp args compatibility |
||
|
Kagura
|
6d9ca9de93 |
fix(daemon): suppress agent terminal windows on Windows (#1474)
* fix(daemon): suppress agent terminal windows on Windows (#1471) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: add hideAgentWindow to detectCLIVersion and avoid SysProcAttr overwrite - Add missing hideAgentWindow(cmd) call in detectCLIVersion (claude.go:554) so --version checks don't flash console windows on Windows. - Refactor hideAgentWindow to preserve existing SysProcAttr fields instead of overwriting the entire struct. --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
LinYushen
|
d97aec83d7 |
fix: pass model to Hermes ACP and add hermes to InjectRuntimeConfig (#1203)
* fix: pass model to Hermes ACP session/new and add hermes to InjectRuntimeConfig - hermes.go: include opts.Model in session/new params so Hermes uses the configured model instead of its default (fixes local LLM failures) - runtime_config.go: add "hermes" to the AGENTS.md provider list so Hermes receives the Multica runtime instructions and skill discovery Fixes: https://github.com/multica-ai/multica/issues/1195 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * fix(hermes): drop false native-skill claim and add regression tests The previous change added 'hermes' to the 'skills discovered automatically' branch of buildMetaSkillContent, but resolveSkillsDir has no Hermes case so skills still land in the .agent_context/skills/ fallback. AGENTS.md ended up claiming native discovery while the files were somewhere else, which would mislead Hermes (and future debuggers). - Move 'hermes' to the fallback branch alongside 'gemini' so AGENTS.md points Hermes at .agent_context/skills/ — matching where writeContextFiles actually writes them. - Extract buildHermesSessionParams so the session/new payload is unit-testable. - Add regression tests covering: * buildHermesSessionParams includes/omits 'model' correctly * InjectRuntimeConfig('hermes') writes AGENTS.md with the fallback hint * writeContextFiles('hermes') writes skills to .agent_context/skills/ Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: CC-Girl <cc-girl@multica.ai> |
||
|
Bohan Jiang
|
4ac43e9e49 |
feat(daemon): log agent invocation at info level (#1428)
Surface the actual exec path + argv for every agent backend at INFO so operators can see the exact command without flipping to debug. Also add the missing log line in pi.go for consistency with the other nine backends. |
||
|
devv-eve
|
9e47b83f02 |
feat(agent): add Kimi CLI as agent runtime (#1400)
* feat(agent): add Kimi CLI as agent runtime
Adds support for Moonshot AI's Kimi Code CLI (https://github.com/MoonshotAI/kimi-cli)
as a new agent runtime, alongside Claude, Codex, OpenCode, OpenClaw, Hermes,
Gemini, Pi, Cursor and Copilot.
Kimi Code CLI implements the standard Agent Client Protocol (ACP) via the
`kimi acp` subcommand, so the new `kimiBackend` reuses the existing
hermesClient JSON-RPC transport in the agent package — only the binary,
client identity, log prefix, and tool-name extraction differ.
Wiring:
- server/pkg/agent: new kimiBackend + kimi_test.go; registered in New(),
LaunchHeader map, and the supported-types coverage test.
- server/internal/daemon/config.go: probes `kimi` (overridable via
MULTICA_KIMI_PATH / MULTICA_KIMI_MODEL).
- server/internal/daemon/execenv: writes AGENTS.md as the runtime context
file (Kimi reads AGENTS.md natively via /init), and writes skills under
`.kimi/skills/` so they are auto-discovered by the project-level skill
loader.
- packages/views/runtimes: ProviderLogo gains a Kimi mark.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* feat(agent/kimi): support per-agent model selection via ACP set_model
Wire Kimi into the model dropdown introduced in #1399:
- ListModels gets a 'kimi' case that drives the same ACP
initialize + session/new handshake as Hermes; both share a new
discoverACPModels helper and parseACPSessionNewModels parser
so future ACP backends only need a small provider entry.
- kimiBackend now issues session/set_model after session/new when
opts.Model is non-empty, mirroring the Hermes flow. Failures
fail the task instead of silently falling back to Kimi's
default model — silent fallback would hide that the dropdown
pick wasn't honoured.
Verified: go build ./..., go test ./pkg/agent/... ./internal/daemon/... ./internal/handler/..., pnpm typecheck and pnpm test (138 passed).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* refactor(agent): address code review feedback on Kimi runtime
- Share ACP provider-error sniffer between hermes and kimi. Previously
only hermes promoted stderr-observed 4xx/5xx into a failed task;
kimi would report "completed + empty output" when the Moonshot
upstream rejected a request (expired token, rate limit, …). Rename
hermesProviderErrorSniffer → acpProviderErrorSniffer and parameterise
the provider name; wire it into kimiBackend.Execute the same way.
- Rename extractHermesSessionID → extractACPSessionID (shared by all
ACP backends) so the name matches parseACPSessionNewModels.
- Drop the redundant second argument to kimiToolNameFromTitle; the
Message struct has only one relevant field (Tool), so passing it
twice was a dead fallback. Document that the function normalises
residual capitalised kimi titles not caught by hermesToolNameFromTitle.
- Remove kimi-only cmd.WaitDelay override; the hermes baseline is
fine for both and divergence adds noise.
- Add TestKimiBackendSetModelFailureFailsTask: fake `kimi acp` binary
that returns a JSON-RPC error for session/set_model, asserts that
the task result surfaces status=failed with the model name + upstream
message and preserves the session id.
- Fix stale agent listings in agent.go / daemon/config.go doc comments
(missing cursor, gemini, copilot).
All: `go build ./...`, `go vet ./...`, `go test ./pkg/agent/...
./internal/daemon/... ./internal/handler/...` green.
* fix(agent/kimi): pass --yolo so Shell tools don't hang on approval
Kimi's default config has `default_yolo = false`. Every Shell/file-mutating
tool call causes kimi acp to send a `session/request_permission` request
and block (up to 300s) waiting for a response. The daemon's hermesClient
only handles `session/update` notifications — permission requests go
unanswered, the tool call times out, and the UI loop eventually dies
("UI loop timed out"). Observed with the first real kimi task: agent sat
as Live for ~7 minutes before the daemon killed it.
The fix mirrors hermes' HERMES_YOLO_MODE=1 override: pass `--yolo` to
`kimi` so it auto-approves everything. `--yolo` is a top-level flag on
the `kimi` CLI (not a flag on `kimi acp`), so it must come before the
`acp` subcommand in argv. Added to kimiBlockedArgs so user custom_args
can't strip it.
While here, fix a related bug that made kimi tool names show up empty
in the daemon log ("tool #1: "): hermesToolNameFromTitle's fallback
returned `kind` when neither title-with-colon nor kind matched a known
tool. Kimi's ACP `tool_call` emits bare titles like "Shell" or "Read
file" with no `kind` at all, so we'd drop the title on the floor before
kimiToolNameFromTitle ever got a chance to map it. Now: preserve the
title when kind is unclassified; hermes titles always carry a colon so
this branch never fires for hermes.
Tests:
- TestKimiBackendPassesYoloFlag — fake binary that records its argv,
asserts --yolo comes before acp.
- TestHermesToolNameFromTitle rows for bare kimi-style titles.
- Existing suite green: go build, go vet, full pkg/agent + daemon +
handler test packages.
* fix(agent/acp): auto-approve session/request_permission from agent
The previous attempt (`kimi --yolo acp`) was a no-op. Inspected the
kimi-cli source: the `acp` Typer subcommand takes no parameters, so
flags on the root `kimi` command are dropped before `acp_main()` runs
— it's impossible to opt into YOLO mode through CLI flags for ACP.
The real fix is on our side: respond to session/request_permission.
ACP is bidirectional. When kimi runs a Shell or file-write tool, it
sends `session/request_permission` (agent → client, JSON-RPC request
with id + method) and waits up to 300s for a response. Our existing
hermesClient.handleLine only dispatched: (id + result/error) →
handleResponse, and (no id + method) → handleNotification. A request
with BOTH id and method fell through and got silently dropped — kimi
timed out, UI loop died, task sat stuck for 7 minutes.
Add handleAgentRequest: for session/request_permission, echo the id
and respond with outcome=selected, optionId=approve_for_session. The
daemon is headless; there's no user to prompt. `approve_for_session`
lets the agent remember the action so subsequent identical calls
(every Shell, every file write) skip the round-trip entirely. For any
other agent → client method, reply with standard -32601 method-not-
found so the agent doesn't block.
Also:
- Add writeMu so request() (main goroutine) and handleAgentRequest
(reader goroutine) don't interleave JSON frames on stdin.
- Revert the `--yolo acp` flag — it's a no-op, and carrying it in
kimiBlockedArgs gives the wrong impression that it does something.
Comment in kimi.go now points at handleAgentRequest as the real fix.
Tests:
- TestHermesClientAutoApprovesPermissionRequest: inject a
session/request_permission, assert the reply echoes the id and
carries {outcome: selected, optionId: approve_for_session}.
- TestHermesClientReplesMethodNotFoundForUnknownAgentRequest: confirm
unknown agent → client methods get JSON-RPC -32601 instead of silence.
- TestKimiBackendInvokesACPSubcommand replaces the yolo-flag assertion
with a negative assertion: no dead --yolo / --auto-approve / -y on
argv, since they'd pretend to do something they can't.
All: go build ./..., go vet ./..., go test ./pkg/agent/... green.
* fix(agent/acp): surface kimi tool input/output via content blocks
Kimi-cli emits tool_call and tool_call_update ACP frames with the
input/output inside a `content` array of ContentToolCallContent
blocks (shape: {type:"content", content:{type:"text", text:"..."}}),
not in the hermes-style `rawInput` map / `rawOutput` string. Our
parser only looked at rawInput/rawOutput, so the daemon recorded
empty Input and Output for every kimi tool — the execution-history
UI showed blank terminal panels even for commands that ran fine.
Add extractACPToolCallText() and a fallback in handleToolCallStart /
handleToolCallUpdate: when rawInput is nil / rawOutput is empty, pull
the text out of the content blocks. rawInput / rawOutput still take
precedence so hermes' behaviour is untouched. Terminal /
FileEditToolCallContent blocks are skipped (we have nothing to render
them as — kimi only emits TerminalToolCallContent when the client
advertises terminal capability, which we don't).
Tests:
- TestHermesClientHandleToolCallStartKimiContent — content array →
Input.text populated.
- TestHermesClientHandleToolCallCompleteKimiContent — multi-block
content → Output concatenated with newline separator.
- TestHermesClientHandleToolCallRawOutputTakesPrecedence — hermes
rawOutput still wins when both are present.
- TestExtractACPToolCallText — unit coverage for the helper
(single/multiple text blocks, terminal-block skip, empty input).
* fix(agent/acp): buffer streaming tool args so Input isn't empty in UI
kimi-cli streams tool args token-by-token via tool_call_update frames
— the initial tool_call carries an empty content block and each
subsequent in_progress update carries the cumulative JSON so far
(`{`, `{"comma`, `{"command": "echo`, …). The final completed update
then carries the tool's stdout, not the args. Observed per kimi-cli
acp/session.py::_send_tool_call{,_part,_result} and confirmed by
driving a real Shell call end-to-end: 10 in_progress frames, last
with `{"command": "echo hello world"}`, then completed with `hello
world\n`.
Our previous handleToolCallStart emitted MessageToolUse on the first
tool_call frame, capturing the empty content — so every kimi tool
appeared in the execution-history UI with a blank input. Output was
correct (fix
|
||
|
Jiayuan Zhang
|
b291db11c2 |
feat(agents): add per-agent model field with provider-aware dropdown (#1399)
Adds a first-class `model` field on agents so users can pick the LLM model from the create / settings UI instead of editing `custom_env` / `custom_args`. Each provider's dropdown is populated from the live CLI when possible (`opencode models`, `pi --list-models`, `openclaw agents list --json`, `cursor-agent --list-models`, hermes ACP `session/new` → `SessionModelState`), with a static catalog for providers that don't enumerate.
Daemon resolves the runtime model as `agent.model → MULTICA_<PROVIDER>_MODEL → ""` — empty passes through so each backend's CLI picks its own default, avoiding static-guess drift.
Per-provider honouring:
- Claude / Codex / OpenCode / Cursor / Gemini / Pi / Copilot — CLI `--model` / thread payload.
- OpenClaw — `opts.Model` is mapped to `--agent <name>` (the CLI rejects `--model`).
- Hermes — `session/set_model` ACP RPC; stderr is sniffed for provider-level errors so HTTP 4xx from the configured LLM surfaces instead of "empty output"; explicit-model failures mark the task `failed`.
Supporting changes: migration 050 adds `agent.model`; daemon ↔ server heartbeat piggyback carries a model-discovery request; new REST endpoints under `/api/runtimes/{id}/models`; `multica agent create --model` / `update --model`; shared `ModelDropdown` in `packages/views/agents` (searchable, creatable, provider-grouped, default-badge, runtime-supported gate).
|
||
|
Junghwan
|
7395b51aee |
fix(agent): apply filterCustomArgs to hermes backend for parity (#1122)
Every other backend (claude, codex, opencode, openclaw, gemini) filters opts.CustomArgs through a per-backend blocked map so protocol-critical flags can't be overridden via the Create Agent UI. The hermes backend appended CustomArgs directly to argv, so any future flag we add to the map would be silently bypassed here. Add hermesBlockedArgs (with 'acp' as the pinned subcommand) and route CustomArgs through filterCustomArgs. Behaviour is identical for today's use cases; the change prevents accidental protocol-flag overrides and brings hermes in line with the other five backends. Closes #1113 Co-authored-by: shaun0927 <shaun0927@users.noreply.github.com> |
||
|
Bohan Jiang
|
b6d30c0e00 |
feat(agent): log full command line at debug level when spawning agents (#1071)
Add a debug-level log line in every agent backend (claude, codex, opencode, openclaw, gemini, hermes) that prints the executable path and full argument list when spawning the agent process. Helps diagnose custom args, model overrides, and other CLI flag issues. |
||
|
Bohan Jiang
|
ce447c7f06 |
feat(agent): add custom CLI arguments support (#986)
* feat(agent): add custom CLI arguments support Allow users to configure custom CLI arguments per agent that get appended to the agent subprocess command at launch time. This enables use cases like specifying different models (--model o3), max turns, or other provider-specific flags without needing separate runtimes. Changes: - Add custom_args JSONB column to agent table (migration 041) - Update API handler to accept/return custom_args in create/update - Pass custom_args through claim endpoint to daemon - Append custom_args to CLI commands for all agent backends - Add ExecOptions.CustomArgs field in agent package - Add Custom Args tab in agent detail UI - Add --custom-args flag to CLI agent create/update commands Closes MUL-802 * fix(agent): filter protocol-critical flags from custom_args Add per-backend filtering of custom_args to prevent users from accidentally overriding flags that the daemon hardcodes for its communication protocol (e.g. --output-format, --input-format, --permission-mode for Claude). This follows the same pattern as custom_env's isBlockedEnvKey: we only block the small, stable set of flags that would break the daemon↔agent protocol — not every possible dangerous flag. Workspace members are trusted for everything else. Each backend defines its own blocked set: - Claude: -p, --output-format, --input-format, --permission-mode - Gemini: -p, --yolo, -o - Codex: --listen - OpenCode: --format - OpenClaw: --local, --json, --session-id, --message - Hermes: none (ACP is positional) Includes unit tests for the filtering logic. * fix(agent): address code review nits for custom_args - Replace module-level `nextArgId` counter with `crypto.randomUUID()` in custom-args-tab.tsx to avoid SSR ID conflicts - Add unit tests for custom args passthrough and blocked-arg filtering in both Claude and Gemini arg builders |
||
|
LinYushen
|
a25886102a |
feat(agent): add Hermes Agent Provider via ACP protocol (#623)
* feat(agent): add Hermes Agent Provider via ACP protocol Integrate Hermes as a new agent backend using the ACP (Agent Communication Protocol) JSON-RPC 2.0 over stdio — the same pattern as the Codex provider but with ACP-specific methods. - New hermesBackend spawns `hermes acp` and drives initialize → session/new → session/prompt lifecycle - Handles session/update notifications: agent_message_chunk, agent_thought_chunk, tool_call, tool_call_update, usage_update - Auto-approves tool executions via HERMES_YOLO_MODE env var - Supports session resume, model override, system prompt injection - Token usage extracted from PromptResponse and usage_update events - Auto-detected at daemon startup via MULTICA_HERMES_PATH env var Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat(ui): optimize runtime icons and fix create-agent dialog overflow - Replace OpenClaw pixel-art icon (32 rects) with clean vector paths - Add Hermes provider icon (NousResearch mascot, 48x48 webp data URI) - Use provider-specific icons in runtime selector instead of generic Monitor - Fix dialog overflow: add min-w-0 to grid item so truncate works Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(agent): add required mcpServers param to Hermes ACP session/new ACP SDK v0.11.2 requires mcpServers as a mandatory field in NewSessionRequest. Without it, Pydantic validation fails with "Invalid params" and the agent immediately errors out. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |