Files
multica/server/pkg/db/queries/agent.sql
Bohan Jiang 2bec2221d2 feat(agent): per-agent thinking_level for claude + codex (MUL-2339) (#2865)
* feat(agent): persist thinking_level per agent (MUL-2339)

Adds a nullable `thinking_level` column to the `agent` table so the
backend can route a runtime-native reasoning/effort token (e.g. Claude's
`xhigh`, Codex's `minimal`) through to the agent CLI on every dispatch.

The column is intentionally TEXT rather than an enum — Claude and Codex
publish overlapping but distinct vocabularies and we want the persisted
value to round-trip exactly through whichever CLI receives it. NULL is
the "use runtime default" sentinel that every downstream consumer reads
as "do not inject --effort / reasoning_effort".

This commit is just the storage layer (migration + sqlc); subsequent
commits wire it through the API, daemon, and agent backends.

Co-authored-by: multica-agent <github@multica.ai>

* feat(agent-backend): inject reasoning effort for claude + codex (MUL-2339)

Extends ExecOptions with a runtime-native ThinkingLevel string and wires
it into the Claude and Codex backends. Discovery is driven by the local
CLI so the daemon advertises whatever the host install supports rather
than a hand-maintained list that goes stale.

Per Elon's PR1 review:
- Claude: parses `claude --help` to learn the `--effort` superset and
  projects through a per-model allow-list (xhigh is Opus-only; max is
  session-only on the smaller models). Falls back to a conservative
  static list when the binary is missing or help drift hides the line.
- Codex: drives `codex debug models --output json` so per-model
  reasoning subsets and the documented default come directly from the
  CLI. The older config-error probe trick is gone — the JSON path is
  stable and doesn't pollute stderr with an intentional misconfig.
- Cache key includes (provider, executablePath, cliVersion) so a CLI
  upgrade invalidates entries that referenced the older help / catalog.

Per Trump's PR1 constraint, all three Codex injection points
(thread/start.config, thread/resume.config, turn/start.effort) flow
through one helper (`applyCodexReasoningEffort`) so they cannot drift
independently. The shared `codexReasoningCases` fixture in
`thinking_test.go` asserts the same value→{shape, key} contract at
each site for every level the runtimes know about.

Claude's `--effort` is also added to `claudeBlockedArgs` so a user
custom_args entry can't silently outvote the daemon-injected value.

Co-authored-by: multica-agent <github@multica.ai>

* feat(api): wire thinking_level through API + daemon contract (MUL-2339)

End-to-end plumbing for the per-agent reasoning/effort setting:

- AgentResponse / TaskAgentData now carry `thinking_level`; the daemon's
  claim response includes it and the daemon's executor passes it through
  to agent.ExecOptions, where the Claude and Codex backends already know
  what to do with it.
- ModelEntry on the runtime-models wire format gains a `thinking` block
  carrying `supported_levels` + `default_level` per model so the UI can
  render a runtime-aware picker without the server having to know about
  the local CLI install. `handleModelList` projects the agent-package
  catalog (including the new Thinking field) into the wire shape.
- CreateAgent / UpdateAgent gate the field with a synchronous provider
  enum check (claude / codex only today). UpdateAgent is tri-state:
  field omitted = no change, "" = explicit clear (new
  `ClearAgentThinkingLevel` query, mirrors the existing mcp_config null
  pattern), non-empty = validate then set.

Per Trump's PR1 review, the API NEVER auto-clears on a runtime/model
swap and ALWAYS returns 400 on an unknown literal value — same shape
across CreateAgent, UpdateAgent, and combined patches that move
runtime + level in one request. Per-model combination failures (e.g.
`xhigh` against a model that only supports up to `high`) surface as a
daemon-side task error, not a silent server-side rewrite.

TS types follow the same shape: `Agent.thinking_level`,
`CreateAgentRequest`/`UpdateAgentRequest` add the field, `RuntimeModel`
grows a `thinking` block. Older backends omit the field, which the
front-end treats as "no picker for this model" — installed desktop
builds keep working.

Co-authored-by: multica-agent <github@multica.ai>

* fix(agent): correct codex debug models argv + pin via runner test (MUL-2339)

`codex debug models --output json` is rejected by codex-cli 0.131.0 —
the subcommand emits JSON on stdout by default and has no `--output`
flag. Drop the flag and add `--bundled` to skip the network refresh
discovery doesn't need. Move the argv to a package-level var and add
a test that runs a fake `codex` to assert the binary actually
receives exactly `debug models --bundled`, so the contract can't
silently drift on the next refactor.

Also teach ValidateThinkingLevel to resolve an empty model to the
provider's default model entry. Without this, every default-model
task with a persisted thinking_level would be misjudged "unknown
model" by the daemon guard.

Co-authored-by: multica-agent <github@multica.ai>

* fix(api): reject runtime switch that would leave invalid thinking_level (MUL-2339)

A PATCH that changed `runtime_id` without touching `thinking_level`
used to silently keep the existing value, so a Claude agent storing
`max` could land on a Codex runtime where `max` is not a recognised
token at all, and the daemon would receive a literal-invalid level.

Hold the same "always 400 on literal-invalid, never silent coerce"
rule on this implicit path. When runtime_id changes and the existing
value is not in the new provider's enum, return 400 with the
recovery options (clear via `thinking_level=""` or re-set in the
same PATCH).

Add coverage for both the kept-when-still-valid and the rejected
cases, plus the two recovery paths (clear and replace).

Co-authored-by: multica-agent <github@multica.ai>

* fix(daemon): guard runTask with per-model thinking_level validator (MUL-2339)

ValidateThinkingLevel existed but had no call site — `task.Agent.
ThinkingLevel` flowed straight into ExecOptions, so `xhigh` configured
on a non-Opus Claude model, or API-side stale values that escaped the
provider enum gate, would be injected anyway.

Run the validator before building ExecOptions. Invalid combinations
log a warning and drop the level instead of failing the task: the
agent still runs, just at the runtime's default reasoning effort.
Discovery errors fail open (keep the level, let the CLI surface any
objection) so a transient `claude --help` failure can't strand work.

Empty model is forwarded as-is; the validator resolves it to the
provider's default model internally per the cross-package contract.

Co-authored-by: multica-agent <github@multica.ai>

* chore(agent): drop stale `--output json` comments + unused scanner (MUL-2339)

Codex CLI's `debug models` subcommand emits JSON without an `--output`
flag, and `parseCodexDebugModels` never read from the bufio.Scanner.
Sync the comments with the actual invocation and remove the dead init.

Co-authored-by: multica-agent <github@multica.ai>

---------

Co-authored-by: multica-agent <github@multica.ai>
2026-05-20 12:30:10 +08:00

551 lines
24 KiB
SQL

-- name: ListAgents :many
SELECT * FROM agent
WHERE workspace_id = $1 AND archived_at IS NULL
ORDER BY created_at ASC;
-- name: ListAllAgents :many
SELECT * FROM agent
WHERE workspace_id = $1
ORDER BY created_at ASC;
-- name: GetAgent :one
SELECT * FROM agent
WHERE id = $1;
-- name: GetAgentInWorkspace :one
SELECT * FROM agent
WHERE id = $1 AND workspace_id = $2;
-- name: CreateAgent :one
INSERT INTO agent (
workspace_id, name, description, avatar_url, runtime_mode,
runtime_config, runtime_id, visibility, max_concurrent_tasks, owner_id,
instructions, custom_env, custom_args, mcp_config, model, thinking_level
) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16)
RETURNING *;
-- name: UpdateAgent :one
UPDATE agent SET
name = COALESCE(sqlc.narg('name'), name),
description = COALESCE(sqlc.narg('description'), description),
avatar_url = COALESCE(sqlc.narg('avatar_url'), avatar_url),
runtime_config = COALESCE(sqlc.narg('runtime_config'), runtime_config),
runtime_mode = COALESCE(sqlc.narg('runtime_mode'), runtime_mode),
runtime_id = COALESCE(sqlc.narg('runtime_id'), runtime_id),
visibility = COALESCE(sqlc.narg('visibility'), visibility),
status = COALESCE(sqlc.narg('status'), status),
max_concurrent_tasks = COALESCE(sqlc.narg('max_concurrent_tasks'), max_concurrent_tasks),
instructions = COALESCE(sqlc.narg('instructions'), instructions),
custom_env = COALESCE(sqlc.narg('custom_env'), custom_env),
custom_args = COALESCE(sqlc.narg('custom_args'), custom_args),
mcp_config = COALESCE(sqlc.narg('mcp_config'), mcp_config),
model = COALESCE(sqlc.narg('model'), model),
thinking_level = COALESCE(sqlc.narg('thinking_level'), thinking_level),
updated_at = now()
WHERE id = $1
RETURNING *;
-- name: ClearAgentThinkingLevel :one
-- Explicit NULL-clear for thinking_level. COALESCE-based UpdateAgent cannot
-- set the column back to NULL, so the API layer routes "user picked Default"
-- through this dedicated query.
UPDATE agent SET thinking_level = NULL, updated_at = now()
WHERE id = $1
RETURNING *;
-- name: ClearAgentMcpConfig :one
UPDATE agent SET mcp_config = NULL, updated_at = now()
WHERE id = $1
RETURNING *;
-- name: ArchiveAgent :one
UPDATE agent SET archived_at = now(), archived_by = $2, updated_at = now()
WHERE id = $1
RETURNING *;
-- name: ArchiveAgentsByRuntime :many
-- Bulk-archives every active agent bound to any runtime in the given set.
-- Used when revoking a leaving member's runtimes so agents pinned to those
-- runtimes can no longer be assigned new work. Returns the affected rows so
-- the caller can broadcast agent:archived per agent.
UPDATE agent
SET archived_at = now(), archived_by = @archived_by, updated_at = now()
WHERE runtime_id = ANY(@runtime_ids::uuid[]) AND archived_at IS NULL
RETURNING *;
-- name: RestoreAgent :one
UPDATE agent SET archived_at = NULL, archived_by = NULL, updated_at = now()
WHERE id = $1
RETURNING *;
-- name: ListAgentTasks :many
SELECT * FROM agent_task_queue
WHERE agent_id = $1
ORDER BY created_at DESC;
-- name: CreateAgentTask :one
INSERT INTO agent_task_queue (
agent_id, runtime_id, issue_id, status, priority, trigger_comment_id,
trigger_summary, force_fresh_session, is_leader_task
)
VALUES (
$1, $2, $3, 'queued', $4, sqlc.narg(trigger_comment_id),
sqlc.narg(trigger_summary),
COALESCE(sqlc.narg('force_fresh_session')::boolean, FALSE),
COALESCE(sqlc.narg('is_leader_task')::boolean, FALSE)
)
RETURNING *;
-- name: CreateQuickCreateTask :one
-- Quick-create tasks have no issue / chat / autopilot link; the entire job
-- description (prompt, requester, workspace) lives in context JSONB. The
-- daemon detects this variant via context.type == "quick_create".
INSERT INTO agent_task_queue (agent_id, runtime_id, issue_id, status, priority, context)
VALUES ($1, $2, NULL, 'queued', $3, $4)
RETURNING *;
-- name: LinkTaskToIssue :exec
-- Attaches the issue a quick-create task produced back to the task row, once
-- the agent has finished and the issue exists. Guarded by `issue_id IS NULL`
-- so this never overwrites an issue id that was set at task creation (only
-- quick-create tasks land here unset). Fixes the activity row staying on
-- "Creating issue" forever after completion.
UPDATE agent_task_queue
SET issue_id = $2
WHERE id = $1 AND issue_id IS NULL;
-- name: CreateRetryTask :one
-- Clones a parent task into a fresh queued attempt. Carries forward the
-- agent's resume context (session_id/work_dir) so the child can continue
-- the conversation when the backend supports it. attempt is incremented;
-- max_attempts, trigger_comment_id, and is_leader_task are inherited so
-- the retried task keeps the same squad-role provenance as its parent and
-- the self-trigger guard in shouldEnqueueSquadLeaderOnComment continues to
-- recognise it as a leader task.
INSERT INTO agent_task_queue (
agent_id, runtime_id, issue_id, chat_session_id, autopilot_run_id,
status, priority, trigger_comment_id, trigger_summary, context,
session_id, work_dir,
attempt, max_attempts, parent_task_id, is_leader_task
)
SELECT
p.agent_id, p.runtime_id, p.issue_id, p.chat_session_id, p.autopilot_run_id,
'queued', p.priority, p.trigger_comment_id, p.trigger_summary, p.context,
p.session_id, p.work_dir,
p.attempt + 1, p.max_attempts, p.id, p.is_leader_task
FROM agent_task_queue p
WHERE p.id = $1
RETURNING *;
-- name: CancelAgentTasksByIssue :many
-- Cancels every active task on the issue and returns the affected rows so the
-- caller can reconcile each agent's status and broadcast task:cancelled events
-- (#1587). Prior :exec form silently dropped that info, so internal cancel
-- paths (issue status flips to cancelled/done, etc.) left agents stuck at
-- status="working" with no self-correction.
UPDATE agent_task_queue
SET status = 'cancelled', completed_at = now()
WHERE issue_id = $1 AND status IN ('queued', 'dispatched', 'running')
RETURNING *;
-- name: CancelAgentTasksByIssueAndAgent :many
-- Cancels active tasks for a single (issue, agent) pair without touching
-- tasks belonging to other agents on the same issue. Used by the manual
-- rerun flow so re-running the assignee doesn't collateral-cancel a
-- still-running @-mention agent on the same issue.
UPDATE agent_task_queue
SET status = 'cancelled', completed_at = now()
WHERE issue_id = $1 AND agent_id = $2 AND status IN ('queued', 'dispatched', 'running')
RETURNING *;
-- name: CancelAgentTasksByAgent :many
-- Bulk-cancel every active (queued/dispatched/running) task for an agent.
-- Returns the affected rows so callers can broadcast task:cancelled events.
-- Mirrors the shape of CancelAgentTasksByIssue / CancelAgentTasksByIssueAndAgent
-- (also :many + RETURNING + completed_at) so the three sibling cancel paths
-- behave consistently.
UPDATE agent_task_queue
SET status = 'cancelled', completed_at = now()
WHERE agent_id = $1 AND status IN ('queued', 'dispatched', 'running')
RETURNING *;
-- name: CancelAgentTasksByTriggerComment :many
-- Cancels active tasks whose trigger is the given comment. Called when a
-- comment is deleted so the agent does not run with the now-deleted content
-- already embedded in its prompt. Must run BEFORE the comment row is deleted
-- because the FK ON DELETE SET NULL would otherwise nullify trigger_comment_id
-- and we'd lose the ability to find the affected tasks.
UPDATE agent_task_queue
SET status = 'cancelled', completed_at = now()
WHERE trigger_comment_id = $1 AND status IN ('queued', 'dispatched', 'running')
RETURNING *;
-- name: CancelAgentTasksByChatSession :many
-- Cancels active tasks belonging to a chat session. Called from
-- DeleteChatSession so the daemon doesn't keep running work whose result
-- has nowhere to land. Must run BEFORE the chat_session row is deleted —
-- the FK ON DELETE SET NULL would otherwise nullify chat_session_id and we
-- could no longer reach those tasks.
UPDATE agent_task_queue
SET status = 'cancelled', completed_at = now()
WHERE chat_session_id = $1 AND status IN ('queued', 'dispatched', 'running')
RETURNING *;
-- name: GetAgentTask :one
SELECT * FROM agent_task_queue
WHERE id = $1;
-- name: ClaimAgentTask :one
-- Claims the next queued task for an agent, enforcing per-(issue, agent) serialization:
-- a task is only claimable when no other task for the same issue AND same agent is
-- already dispatched or running. This allows different agents to work on the same
-- issue in parallel while preventing a single agent from running duplicate tasks.
-- Chat tasks (issue_id IS NULL) use chat_session_id for serialization instead.
-- Quick-create tasks have no issue / chat / autopilot link, so they serialize on
-- "any other quick-create-shaped task" (all four FKs NULL) for the same agent —
-- otherwise a user mashing the create button could fire concurrent quick-creates
-- whose completion lookup would race over "most recent issue by this agent".
UPDATE agent_task_queue
SET status = 'dispatched', dispatched_at = now()
WHERE id = (
SELECT atq.id FROM agent_task_queue atq
WHERE atq.agent_id = $1 AND atq.status = 'queued'
AND NOT EXISTS (
SELECT 1 FROM agent_task_queue active
WHERE active.agent_id = atq.agent_id
AND active.status IN ('dispatched', 'running')
AND (
(atq.issue_id IS NOT NULL AND active.issue_id = atq.issue_id)
OR (atq.chat_session_id IS NOT NULL AND active.chat_session_id = atq.chat_session_id)
OR (
atq.issue_id IS NULL
AND atq.chat_session_id IS NULL
AND atq.autopilot_run_id IS NULL
AND active.issue_id IS NULL
AND active.chat_session_id IS NULL
AND active.autopilot_run_id IS NULL
)
)
)
ORDER BY atq.priority DESC, atq.created_at ASC
LIMIT 1
FOR UPDATE SKIP LOCKED
)
RETURNING *;
-- name: StartAgentTask :one
UPDATE agent_task_queue
SET status = 'running', started_at = now()
WHERE id = $1 AND status = 'dispatched'
RETURNING *;
-- name: CompleteAgentTask :one
UPDATE agent_task_queue
SET status = 'completed', completed_at = now(), result = $2, session_id = $3, work_dir = $4
WHERE id = $1 AND status = 'running'
RETURNING *;
-- name: GetLastTaskSession :one
-- Returns the session_id and work_dir from the most recent task for a given
-- (agent_id, issue_id) pair, used for session resumption on the auto-retry
-- path. We accept both 'completed' and 'failed' tasks: a failed task may
-- have established a real agent session before crashing (orphaned by a
-- daemon restart, runtime offline, or sweeper timeout), and the daemon pins
-- the resume pointer mid-flight via UpdateAgentTaskSession. Without this,
-- an auto-retry of a mid-run failure would silently start a fresh
-- conversation and lose the in-flight context — exactly what MUL-1128's B
-- branch is meant to fix.
--
-- Manual rerun (TaskService.RerunIssue) does NOT take this path: it sets
-- force_fresh_session=true on the new task, and the daemon claim handler
-- skips this lookup entirely. The user already judged the prior output bad;
-- resuming the same conversation would replay a poisoned state.
--
-- Tasks that ended in a known "poisoned" terminal state are also excluded
-- here so even auto-retry does not inherit the bad session. The daemon
-- classifies these failures (iteration_limit, agent_fallback_message,
-- api_invalid_request) when it detects either an agent fallback marker in
-- the output or an upstream API 400 that means the conversation history
-- itself is unprocessable (oversized image, malformed base64, etc.).
--
-- The error-text ILIKE clause is defense-in-depth for the api_invalid_request
-- shape: a legacy row tagged 'agent_error' (pre-MUL-1921), a deploy-window
-- row that the old code wrote between migration and rollout, or a future
-- error format that escapes the daemon classifier all still get filtered
-- here as long as the canonical Anthropic 400 marker is present in the
-- error text. Migration 079 backfills the failure_reason column itself,
-- so observability stays accurate; this clause guarantees session resume
-- never picks up a bad session even when failure_reason hasn't caught up.
SELECT session_id, work_dir, runtime_id FROM agent_task_queue
WHERE agent_id = $1 AND issue_id = $2
AND (
status = 'completed'
OR (
status = 'failed'
AND COALESCE(failure_reason, '') NOT IN ('iteration_limit', 'agent_fallback_message', 'api_invalid_request')
AND NOT (COALESCE(error, '') ILIKE '%400%' AND COALESCE(error, '') ILIKE '%invalid_request_error%')
)
)
AND session_id IS NOT NULL
ORDER BY COALESCE(completed_at, started_at, dispatched_at, created_at) DESC
LIMIT 1;
-- name: FailAgentTask :one
-- Marks a task as failed. session_id and work_dir are merged via COALESCE so
-- if the agent already established a real session before failing (e.g. it
-- crashed mid-conversation, was cancelled, or hit a tool error) the resume
-- pointer is preserved on the task row. The next chat task can then fall
-- back to GetLastChatTaskSession and continue the conversation instead of
-- silently starting over.
--
-- failure_reason is a coarse classifier consumed by the auto-retry path;
-- 'agent_error' is the safe default when the daemon doesn't supply one.
UPDATE agent_task_queue
SET status = 'failed',
completed_at = now(),
error = $2,
failure_reason = COALESCE(sqlc.narg('failure_reason'), 'agent_error'),
session_id = COALESCE(sqlc.narg('session_id'), session_id),
work_dir = COALESCE(sqlc.narg('work_dir'), work_dir)
WHERE id = $1 AND status IN ('dispatched', 'running')
RETURNING *;
-- name: UpdateAgentTaskSession :exec
-- Pins the resume pointer mid-flight so a daemon crash leaves a usable
-- session_id/work_dir on the task row. No-op if the task is no longer
-- in dispatched/running.
UPDATE agent_task_queue
SET session_id = COALESCE(sqlc.narg('session_id'), session_id),
work_dir = COALESCE(sqlc.narg('work_dir'), work_dir)
WHERE id = $1 AND status IN ('dispatched', 'running');
-- name: RecoverOrphanedTasksForRuntime :many
-- Called by the daemon at startup. Atomically fails any dispatched/running
-- task that the prior incarnation of this runtime owned but did not
-- finalize. Returns the failed rows so callers can hand them to the
-- auto-retry path.
UPDATE agent_task_queue
SET status = 'failed',
completed_at = now(),
error = 'daemon restarted while task was in flight',
failure_reason = 'runtime_recovery'
WHERE runtime_id = $1 AND status IN ('dispatched', 'running')
RETURNING *;
-- name: FailStaleTasks :many
-- Fails tasks stuck in dispatched/running beyond the given thresholds.
-- Handles cases where the daemon is alive but the task is orphaned
-- (e.g. agent process hung, daemon failed to report completion).
UPDATE agent_task_queue
SET status = 'failed', completed_at = now(), error = 'task timed out',
failure_reason = 'timeout'
WHERE (status = 'dispatched' AND dispatched_at < now() - make_interval(secs => @dispatch_timeout_secs::double precision))
OR (status = 'running' AND started_at < now() - make_interval(secs => @running_timeout_secs::double precision))
RETURNING *;
-- name: ExpireStaleQueuedTasks :many
-- Fails tasks that have been sitting in 'queued' for longer than the TTL.
-- This is the cleanup arm of the MUL-1899 "queued backlog" fix: even with the
-- new dispatch-time admission gate that refuses to enqueue when the runtime
-- is offline, we still need to drain the historical 87k+ doomed rows and
-- handle edge cases where a runtime goes offline AFTER a task is already
-- queued (the admission check protects new enqueues, not in-flight queue
-- depth).
--
-- Concurrency safety: the daemon's claim path may race with this sweeper to
-- transition the same row out of 'queued'. We protect against that two
-- ways:
-- 1. The CTE selects victims with FOR UPDATE SKIP LOCKED so a row that is
-- currently being claimed (or otherwise locked) is skipped — no lock
-- contention with the dispatch path, and we won't queue up behind it.
-- 2. The outer UPDATE re-checks status='queued' AND the TTL predicate at
-- apply time. If a daemon claimed the row between selection and update
-- (e.g. lock released after the claim transaction commits), the row is
-- already 'dispatched'/'running' and the WHERE clause filters it out
-- so we cannot clobber an in-flight task.
-- Capped via LIMIT inside the CTE so a single sweep tick cannot monopolise
-- the DB when the backlog is large — the sweeper drains the rest on
-- subsequent ticks.
WITH victims AS (
SELECT id FROM agent_task_queue
WHERE status = 'queued'
AND created_at < now() - make_interval(secs => @ttl_secs::double precision)
ORDER BY created_at ASC
LIMIT @max_per_tick::int
FOR UPDATE SKIP LOCKED
)
UPDATE agent_task_queue t
SET status = 'failed',
completed_at = now(),
error = 'task expired in queue',
failure_reason = 'queued_expired'
FROM victims v
WHERE t.id = v.id
AND t.status = 'queued'
AND t.created_at < now() - make_interval(secs => @ttl_secs::double precision)
RETURNING t.*;
-- name: CancelAgentTask :one
UPDATE agent_task_queue
SET status = 'cancelled', completed_at = now()
WHERE id = $1 AND status IN ('queued', 'dispatched', 'running')
RETURNING *;
-- name: CountRunningTasks :one
SELECT count(*) FROM agent_task_queue
WHERE agent_id = $1 AND status IN ('dispatched', 'running');
-- name: HasActiveTaskForIssue :one
-- Returns true if there is any queued, dispatched, or running task for the issue.
SELECT count(*) > 0 AS has_active FROM agent_task_queue
WHERE issue_id = $1 AND status IN ('queued', 'dispatched', 'running');
-- name: HasPendingTaskForIssue :one
-- Returns true if there is a queued or dispatched (but not yet running) task for the issue.
-- Used by the coalescing queue: allow enqueue when a task is running (so
-- the agent picks up new comments on the next cycle) but skip if a pending
-- task already exists (natural dedup).
SELECT count(*) > 0 AS has_pending FROM agent_task_queue
WHERE issue_id = $1 AND status IN ('queued', 'dispatched');
-- name: HasPendingTaskForIssueAndAgent :one
-- Returns true if a specific agent already has a queued or dispatched task
-- for the given issue. Used by @mention trigger dedup.
SELECT count(*) > 0 AS has_pending FROM agent_task_queue
WHERE issue_id = $1 AND agent_id = $2 AND status IN ('queued', 'dispatched');
-- name: GetLatestTaskIsLeaderForIssueAndAgent :one
-- Returns the is_leader_task flag of the agent's most recent task on this
-- issue, or NULL if the agent has never had a task on this issue. Used by
-- the squad-leader self-trigger guard to tell whether the agent's last
-- activity on the issue was in the leader role or the worker role (an
-- agent that holds both roles in a squad would otherwise be skipped by
-- the role-blind authorID == leaderID check).
SELECT is_leader_task FROM agent_task_queue
WHERE issue_id = $1 AND agent_id = $2
ORDER BY created_at DESC
LIMIT 1;
-- name: ListPendingTasksByRuntime :many
SELECT * FROM agent_task_queue
WHERE runtime_id = $1 AND status IN ('queued', 'dispatched')
ORDER BY priority DESC, created_at ASC;
-- name: ListQueuedClaimCandidatesByRuntime :many
-- Returns rows the runtime can attempt to claim. Status is restricted to
-- 'queued' (in contrast to ListPendingTasksByRuntime which also includes
-- 'dispatched') because dispatched rows are by definition already owned
-- and cannot be re-claimed — including them in the candidate list pads
-- the result with rows that always lose the per-(issue, agent) race in
-- ClaimAgentTask, wasting CPU and a SELECT every poll cycle when the
-- runtime is busy on a long-running task. Backed by the partial index
-- idx_agent_task_queue_claim_candidates so the warm path is cheap.
SELECT * FROM agent_task_queue
WHERE runtime_id = $1 AND status = 'queued'
ORDER BY priority DESC, created_at ASC;
-- name: ListActiveTasksByIssue :many
-- Backs the issue-detail "agent live" banner. Includes 'queued' so the
-- banner shows up the moment a task is enqueued — not only after a runtime
-- claims it. The queued window can be long when the runtime is offline or
-- busy on a prior task, and a silent UI during that window looks like the
-- platform never received the trigger.
SELECT * FROM agent_task_queue
WHERE issue_id = $1 AND status IN ('queued', 'dispatched', 'running')
ORDER BY created_at DESC;
-- name: GetWorkspaceAgentRunCounts :many
-- Total task runs per agent over the trailing 30 days, used by the Agents
-- list RUNS column. 30-day window keeps the count meaningful (a long-dormant
-- agent shouldn't show "5,420 runs from 2 years ago") and keeps the scan
-- bounded as the workspace ages.
SELECT
atq.agent_id,
COUNT(*)::int AS run_count
FROM agent_task_queue atq
JOIN agent a ON a.id = atq.agent_id
WHERE a.workspace_id = $1
AND atq.created_at > now() - INTERVAL '30 days'
GROUP BY atq.agent_id;
-- name: GetWorkspaceAgentActivity30d :many
-- Returns per-agent daily activity buckets for the last 30 days. Single
-- workspace-wide read backs both surfaces:
-- - Agents list ACTIVITY column — uses only the trailing 7 buckets
-- - Agent detail "Last 30 days" panel — uses the full 30
-- 30 days contains 7 days, so one fetch + a client-side .slice(-7) wins
-- over fetching twice. Days with no completion produce no row; the
-- front-end zero-fills.
--
-- Anchored on completed_at (not created_at) because the sparkline answers
-- "what did this agent produce?" not "what was queued at it?". A task that's
-- still in flight has no completed_at and contributes nothing here — that's
-- correct: in-flight tasks are surfaced via the live presence indicator,
-- not the historical trend.
SELECT
atq.agent_id,
DATE_TRUNC('day', atq.completed_at)::timestamptz AS bucket,
COUNT(*)::int AS task_count,
COUNT(*) FILTER (WHERE atq.status = 'failed')::int AS failed_count
FROM agent_task_queue atq
JOIN agent a ON a.id = atq.agent_id
WHERE a.workspace_id = $1
AND atq.completed_at IS NOT NULL
AND atq.completed_at > now() - INTERVAL '30 days'
GROUP BY atq.agent_id, bucket
ORDER BY atq.agent_id, bucket;
-- name: ListWorkspaceAgentTaskSnapshot :many
-- Returns the tasks needed to derive each agent's current presence:
-- - All active tasks (queued / dispatched / running) — for working signal + counts
-- - Each agent's most recent OUTCOME task (completed / failed) — for sticky
-- failed signal
-- The front-end picks "active wins, else latest outcome" — see derive-presence.ts.
--
-- Cancelled tasks are excluded from the outcome half on purpose: cancel is a
-- procedural signal ("attempt aborted"), not an outcome. It tells us nothing
-- about whether the agent works, so it must NOT be allowed to mask a prior
-- failure. Concretely: if an agent fails and then the user cancels the queued
-- retry (or the parent issue closes and cascades cancels), the failed signal
-- has to stay red. Only a real success (completed) or a fresh attempt (active)
-- clears it.
--
-- No UI windows in SQL: stickiness is decided by "is the latest outcome a
-- failure?", not a 2-minute clock. JOINs agent because agent_task_queue has
-- no workspace_id column.
SELECT atq.* FROM agent_task_queue atq
JOIN agent a ON a.id = atq.agent_id
WHERE a.workspace_id = $1
AND atq.status IN ('queued', 'dispatched', 'running')
UNION ALL
SELECT t.* FROM (
SELECT DISTINCT ON (atq.agent_id) atq.*
FROM agent_task_queue atq
JOIN agent a ON a.id = atq.agent_id
WHERE a.workspace_id = $1
AND atq.status IN ('completed', 'failed')
ORDER BY atq.agent_id, atq.completed_at DESC NULLS LAST
) t;
-- name: ListTasksByIssue :many
SELECT * FROM agent_task_queue
WHERE issue_id = $1
ORDER BY created_at DESC;
-- name: UpdateAgentStatus :one
UPDATE agent SET status = $2, updated_at = now()
WHERE id = $1
RETURNING *;
-- name: RefreshAgentStatusFromTasks :one
UPDATE agent AS a
SET status = CASE WHEN EXISTS (
SELECT 1 FROM agent_task_queue q
WHERE q.agent_id = a.id AND q.status IN ('dispatched', 'running')
) THEN 'working' ELSE 'idle' END,
updated_at = now()
WHERE a.id = $1
RETURNING *;