Files
multica/server/internal/daemon/prompt_test.go
Bohan Jiang e45a8f6c12 fix(daemon): scan comment roots before bulk reads in agent catch-up (MUL-5372) (#6093)
* fix(daemon): scan comment roots before bulk reads in agent catch-up

The mandatory step-3 catch-up in the issue runtime brief asked for
`--recent 10`. `--recent N` caps THREADS, not comments: each returned
thread carries its root plus every descendant with no depth bound, so on
an issue with fewer than N root threads it returns the entire comment
history. Because the step is mandatory and fires on every run, every
reply turn re-read the whole issue -- and on comment-triggered turns it
duplicated the bounded thread read the per-turn message had already
pointed at, so the same bytes were fetched twice.

Lead the step with `--roots-only --summary` instead: every top-level
thread with reply_count and last_activity_at, contents clipped. That
keeps the property the step exists for -- the agent still sees every
thread that exists, so it cannot act on stale context -- and makes the
drill-down into `--thread <id> --tail 30` explicit. `--recent 10` stays
documented for when several complete threads really are needed, now with
its saturation semantics spelled out.

Measured on a live 2-thread issue: 21,249 -> 1,518 bytes for the
mandatory read (-93%), and the duplicate 11,082-byte thread read is gone.

The brief stays byte-identical across runs of a session (MUL-5377): the
new text interpolates only the issue id, no per-run state. The three
per-turn pointers that express the same rule move with it so the two
layers cannot drift.

MUL-5372

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

* refactor(daemon): keep comment-read flag semantics in one place

The previous commit fixed the payload shape but restated the read surface
in four places: the workflow step, both per-turn prompt fallbacks, and the
cold-start hint each explained what `--recent 10` does. `## Available
Commands` is already the brief's single discovery point for these flags,
and `TestInjectRuntimeConfigStaticCatchUp` pins it as such -- so those
restatements were duplicated reference text, and the per-turn ones were
paid on every turn rather than once in the cached prefix.

Move the `--recent N` saturation warning into the `comment list` line in
Available Commands, next to the flags it qualifies, and add `--roots-only`
and `--summary` to that signature so the bounding options are discoverable
where an agent already looks. Workflow steps and per-turn hints now name
only the reads they actually want run.

Per-turn prompt sizes: assignment 1170 -> 749 bytes (-36%), cold-start
comment turn 1550 -> 1355 (-13%). Step 3 is 1065 bytes and no longer
carries a ready-to-paste bulk read.

MUL-5372

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

* docs(daemon): address review nits on comment-catchup change

Three cosmetic follow-ups from review:

- `--recent N` saturation warning said it hands back "the entire
  history"; resolved threads are still folded by default on that read, so
  say so.
- Rename two tests whose names still advertised `--recent` after their
  assertions stopped mentioning it, plus the one added in this branch
  whose name referenced a bulk read the step no longer contains:
  MentionsRecent -> ScansRootsFirst, ScansRootsBeforeBulkRead ->
  ScansRootsFirst.
- Fix the stale doc comment that still described the mandatory read as
  bounded to "the recent active-thread window".

No behavior change.

MUL-5372

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

---------

Co-authored-by: Bohan-J <bohan@devv.ai>
Co-authored-by: multica-agent <github@multica.ai>
2026-07-29 16:08:55 +08:00

1229 lines
48 KiB
Go

package daemon
import (
"strings"
"testing"
"github.com/multica-ai/multica/server/internal/daemon/execenv"
)
// TestBuildQuickCreatePromptRules locks in the rules that govern how the
// quick-create agent is allowed to translate raw user input into the issue
// description body. Each substring corresponds to a concrete failure mode
// observed in production output:
// - meta-instructions ("create an issue", "cc @X") leaking into the body
// - the Context section being misused as an apology log when no external
// references were actually fetched
// - hard-line rules being silently dropped on prompt rewrites
func TestBuildQuickCreatePromptRules(t *testing.T) {
out := buildQuickCreatePrompt(Task{QuickCreatePrompt: "fix the login button color"})
mustContain := []string{
// high-fidelity invariant
"Faithfully restate what the user wants",
"Preserve specific names, identifiers, file paths",
// strip non-spec material: verbal routing wrappers + conversational fillers
"verbal routing wrappers about creating the issue",
"pure conversational fillers",
// cc routing must survive: mention link stays in description so the
// auto-subscribe path fires (multica issue create has no --subscriber flag)
"CC exception",
"auto-subscribes members",
// context section is conditional and must not be an apology log
"include ONLY when the input cited external resources",
"never use it as an apology log",
// output/reporting must be workspace-prefix agnostic. Workspaces can
// use custom issue prefixes, so a successful issue creation should
// not look failed merely because the identifier does not match one
// fixed prefix.
"multica issue create --output json",
"JSON response",
"identifier",
"Do not scrape human output",
"do not assume any workspace issue prefix",
"Created <identifier-or-id>: <title>",
// hard rules
"never invent requirements",
"never reduce multi-sentence input",
}
for _, s := range mustContain {
if !strings.Contains(out, s) {
t.Errorf("buildQuickCreatePrompt output missing required rule: %q", s)
}
}
}
// TestBuildQuickCreatePromptAssigneeIncludesSquads locks in the MUL-2165
// fix: the assignee-resolution rules must tell the agent to consult the
// squad list alongside members and agents. Before this, a quick-create
// input like "assign to <SquadName>" silently fell through to
// "Unrecognized assignee" because squads were never queried.
func TestBuildQuickCreatePromptAssigneeIncludesSquads(t *testing.T) {
out := buildQuickCreatePrompt(Task{QuickCreatePrompt: "fix the login button color"})
mustContain := []string{
"multica squad list",
"Squads are first-class assignees",
"Treat bare @-routing as an assignee directive",
"让 @独立团 review 这个 PR",
"pass the squad's `id` as `--assignee-id`",
}
for _, s := range mustContain {
if !strings.Contains(out, s) {
t.Errorf("buildQuickCreatePrompt assignee block missing %q\n--- output ---\n%s", s, out)
}
}
}
// TestBuildQuickCreatePromptSquadDefaultsToSquad locks in the MUL-2203
// fix: when the picker was a squad, the task runs on the squad's leader
// agent, but the default assignee for issues created by this run must
// point at the SQUAD's UUID — not the leader agent's UUID. The previous
// "default to YOURSELF" instruction made squad-created issues land under
// the leader, hiding them from the squad's delegation flow.
func TestBuildQuickCreatePromptSquadDefaultsToSquad(t *testing.T) {
const (
squadID = "aaaa1111-2222-3333-4444-555555555555"
squadName = "独立团"
leaderID = "bbbb1111-2222-3333-4444-666666666666"
)
out := buildQuickCreatePrompt(Task{
QuickCreatePrompt: "fix the login button color",
Agent: &AgentData{ID: leaderID, Name: "leader-agent"},
SquadID: squadID,
SquadName: squadName,
})
// The default-assignee instruction must point at the squad UUID.
if !strings.Contains(out, "--assignee-id \""+squadID+"\"") {
t.Errorf("buildQuickCreatePrompt with SquadID must default to the squad's UUID, got:\n%s", out)
}
// And it must NOT tell the agent to default to itself (the leader).
if strings.Contains(out, "--assignee-id \""+leaderID+"\"") {
t.Errorf("buildQuickCreatePrompt with SquadID must NOT default to the leader agent's UUID, got:\n%s", out)
}
// The squad name should appear in the instruction so the agent has
// human-readable context for the routing decision.
if !strings.Contains(out, squadName) {
t.Errorf("buildQuickCreatePrompt with SquadID should mention the squad name %q, got:\n%s", squadName, out)
}
// And the prompt must explicitly call out the squad-vs-leader rule
// so the agent does not silently regress to "default to YOURSELF".
mustContain := []string{
"picker SQUAD",
"running on the squad's behalf",
"do not assign it to your own agent UUID",
}
for _, s := range mustContain {
if !strings.Contains(out, s) {
t.Errorf("buildQuickCreatePrompt with SquadID missing %q\n--- output ---\n%s", s, out)
}
}
}
// TestBuildQuickCreatePromptProjectPinning verifies that when the user
// pins a project in the quick-create modal, the prompt instructs the agent
// to pass `--project <uuid>` exactly. Without this, the agent would re-read
// the workspace default and silently drop the user's selection — the same
// "I have to retype 'in project X' every time" failure mode the modal
// addition was meant to fix.
func TestBuildQuickCreatePromptProjectPinning(t *testing.T) {
const projectID = "11111111-2222-3333-4444-555555555555"
out := buildQuickCreatePrompt(Task{
QuickCreatePrompt: "fix the login button color",
ProjectID: projectID,
ProjectTitle: "Web App",
})
mustContain := []string{
"--project \"" + projectID + "\"",
"Web App",
"modal selection is authoritative",
}
for _, s := range mustContain {
if !strings.Contains(out, s) {
t.Errorf("buildQuickCreatePrompt with project missing %q\n--- output ---\n%s", s, out)
}
}
// Without a project, the prompt must keep the legacy "omit" instruction
// so the agent doesn't accidentally start passing --project on plain
// quick-create runs.
plain := buildQuickCreatePrompt(Task{QuickCreatePrompt: "fix the login button color"})
if !strings.Contains(plain, "**project**: omit") {
t.Errorf("buildQuickCreatePrompt without project must keep the omit instruction, got:\n%s", plain)
}
if strings.Contains(plain, "--project") {
t.Errorf("buildQuickCreatePrompt without project must NOT mention --project, got:\n%s", plain)
}
}
func TestBuildQuickCreatePromptExplicitPriorityAndDueDate(t *testing.T) {
out := buildQuickCreatePrompt(Task{
QuickCreatePrompt: "fix the login button color",
QuickCreatePriority: "urgent",
QuickCreateDueDate: "2026-08-01",
})
for _, want := range []string{
"--priority urgent",
"--due-date 2026-08-01",
"quick-create selection is authoritative",
} {
if !strings.Contains(out, want) {
t.Errorf("buildQuickCreatePrompt with explicit fields missing %q\n--- output ---\n%s", want, out)
}
}
if strings.Contains(out, "Map P0/P1") {
t.Errorf("explicit priority must replace inference rules, got:\n%s", out)
}
}
// TestBuildQuickCreatePromptParentPinning verifies that when the user
// opened quick-create from "Add sub issue" on an existing issue, the prompt
// instructs the agent to pass `--parent <uuid>` so the new issue is filed
// as a sub-issue. The frontend already seeds parent_issue_id silently
// through the manual→agent switch, so this is the last hop that has to
// hold up — without the prompt instruction the agent would create a
// standalone issue and the sub-issue relationship would be silently
// dropped.
func TestBuildQuickCreatePromptParentPinning(t *testing.T) {
const (
parentID = "33333333-2222-1111-4444-555555555555"
parentIdentifier = "MUL-2534"
)
out := buildQuickCreatePrompt(Task{
QuickCreatePrompt: "fix the login button color",
ParentIssueID: parentID,
ParentIssueIdentifier: parentIdentifier,
})
mustContain := []string{
"--parent \"" + parentID + "\"",
parentIdentifier,
"modal entry point is authoritative",
"filed as a sub-issue",
}
for _, s := range mustContain {
if !strings.Contains(out, s) {
t.Errorf("buildQuickCreatePrompt with parent missing %q\n--- output ---\n%s", s, out)
}
}
// When only the UUID is available (identifier lookup failed on claim),
// the agent must still get the --parent instruction so the sub-issue
// intent isn't silently dropped.
uuidOnly := buildQuickCreatePrompt(Task{
QuickCreatePrompt: "fix the login button color",
ParentIssueID: parentID,
})
if !strings.Contains(uuidOnly, "--parent \""+parentID+"\"") {
t.Errorf("buildQuickCreatePrompt with parent UUID only must still pin --parent, got:\n%s", uuidOnly)
}
// Without a parent, the prompt must NOT mention --parent at all — a
// plain quick-create run should not start filing sub-issues.
plain := buildQuickCreatePrompt(Task{QuickCreatePrompt: "fix the login button color"})
if strings.Contains(plain, "--parent") {
t.Errorf("buildQuickCreatePrompt without parent must NOT mention --parent, got:\n%s", plain)
}
}
// TestBuildPromptSquadLeaderNoActionForMemberTrigger verifies that the
// squad leader no_action prohibition is injected in the per-turn prompt
// regardless of whether the triggering comment was posted by an agent or
// a member. This was the root cause of the "LGTM is a pure acknowledgment
// — no reply needed. Exiting silently." noise comment: the prohibition
// only fired for agent-triggered comments, so member-triggered ones
// (like "LGTM") bypassed it.
func TestBuildPromptSquadLeaderNoActionForMemberTrigger(t *testing.T) {
task := Task{
IssueID: "issue-123",
TriggerCommentID: "comment-456",
TriggerCommentContent: "LGTM",
TriggerAuthorType: "member",
TriggerAuthorName: "Bohan",
Agent: &AgentData{
Instructions: "Some instructions\n\n## Squad Operating Protocol\n\nYou are the LEADER...",
},
}
out := BuildPrompt(task, "claude")
if !strings.Contains(out, "Squad leader no_action rule") {
t.Errorf("buildCommentPrompt must inject squad leader no_action rule for member-triggered comments, got:\n%s", out)
}
if !strings.Contains(out, "DO NOT post any comment") {
t.Errorf("buildCommentPrompt must contain DO NOT post prohibition for member-triggered squad leader, got:\n%s", out)
}
}
// TestBuildPromptSquadLeaderNoActionForAgentTrigger verifies the rule also
// fires for agent-triggered comments (the original path that already worked).
func TestBuildPromptSquadLeaderNoActionForAgentTrigger(t *testing.T) {
task := Task{
IssueID: "issue-123",
TriggerCommentID: "comment-456",
TriggerCommentContent: "Deploy complete.",
TriggerAuthorType: "agent",
TriggerAuthorName: "deploy-boy",
Agent: &AgentData{
Instructions: "Some instructions\n\n## Squad Operating Protocol\n\nYou are the LEADER...",
},
}
out := BuildPrompt(task, "claude")
if !strings.Contains(out, "Squad leader no_action rule") {
t.Errorf("buildCommentPrompt must inject squad leader no_action rule for agent-triggered comments, got:\n%s", out)
}
}
func TestBuildChatPromptAttachmentIDsCanBeBoundToCreatedIssues(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "please create an issue with this screenshot",
ChatMessageAttachments: []ChatAttachmentMeta{
{ID: "019ec09d-6222-722b-bdfa-427b105d80be", Filename: "shot.png", ContentType: "image/png"},
},
}
out := BuildPrompt(task, "claude")
for _, want := range []string{
"Attachments on this message:",
"id=019ec09d-6222-722b-bdfa-427b105d80be",
"multica attachment download <id>",
"--attachment-id <id>",
} {
if !strings.Contains(out, want) {
t.Errorf("chat prompt missing %q\n--- output ---\n%s", want, out)
}
}
}
func TestBuildChatPromptChannelAwareness(t *testing.T) {
t.Run("slack-backed prompt teaches both read commands", func(t *testing.T) {
out := buildChatPrompt(Task{
ChatSessionID: "sess-1",
ChatChannelType: "slack",
ChatMessage: "你刚刚和 xxx 聊了什么",
})
for _, want := range []string{"Slack", "NOT in Multica", "multica chat history", "multica chat thread", "Do NOT narrate"} {
if !strings.Contains(out, want) {
t.Fatalf("slack-backed prompt missing %q\n--- output ---\n%s", want, out)
}
}
})
t.Run("top-level mention starts with history", func(t *testing.T) {
out := buildChatPrompt(Task{ChatSessionID: "s", ChatChannelType: "slack", ChatInThread: false, ChatMessage: "hi"})
if !strings.Contains(out, "top level: start with `multica chat history`") {
t.Fatalf("expected top-level guidance, got:\n%s", out)
}
})
t.Run("in-thread mention starts with thread", func(t *testing.T) {
out := buildChatPrompt(Task{ChatSessionID: "s", ChatChannelType: "slack", ChatInThread: true, ChatMessage: "hi"})
if !strings.Contains(out, "inside a thread: start with `multica chat thread`") {
t.Fatalf("expected in-thread guidance, got:\n%s", out)
}
})
t.Run("web-only session has no channel block", func(t *testing.T) {
out := buildChatPrompt(Task{
ChatSessionID: "sess-1",
ChatMessage: "hi",
})
if strings.Contains(out, "multica chat history") {
t.Fatalf("web-only chat prompt should not mention channel history, got:\n%s", out)
}
})
}
// TestBuildChatPromptNoNarrationOnEveryChannel pins the THIRD axis of the chat
// channel policy: the no-narration delivery rule keys off "is there a channel at
// all", like the upload axis and unlike the Slack-only history axis.
//
// Regression guard for GH #6006. #4776 introduced the rule for every channel;
// the MUL-4899 split moved it into the Slack branch along with the read commands
// its wording happened to mention, so Feishu/Lark replies silently went back to
// carrying interim narration. The two-layer matrix below could not catch that —
// it only ever asserted the rule on the Slack case.
//
// The carve-out is pinned alongside the prohibition on purpose. A rule phrased
// as "don't say what you just did" reads as forbidding "已创建 Issue X" — the
// actual deliverable for a do-this request — so the two must move together: the
// prohibition covers PROCESS, never the completed outcome.
func TestBuildChatPromptNoNarrationOnEveryChannel(t *testing.T) {
const (
prohibition = "Do NOT narrate planned or in-progress steps"
carveOut = "completed actions are part of the outcome"
)
for _, tc := range []struct {
name string
channelType string
want bool
}{
{name: "slack", channelType: execenv.ChannelTypeSlack, want: true},
{name: "feishu", channelType: execenv.ChannelTypeFeishu, want: true},
{name: "direct chat has no channel to deliver into", channelType: "", want: false},
} {
t.Run(tc.name, func(t *testing.T) {
out := buildChatPrompt(Task{
ChatSessionID: "sess-1",
ChatChannelType: tc.channelType,
ChatMessage: "hi",
})
for _, phrase := range []string{prohibition, carveOut} {
if got := strings.Contains(out, phrase); got != tc.want {
t.Errorf("%q present=%v, want %v\n--- output ---\n%s", phrase, got, tc.want, out)
}
}
if !tc.want {
return
}
// The prohibition must not read as a blanket ban on past tense. If a
// future edit drops the carve-out, an agent asked to create an issue
// has no way left to report that it did.
if strings.Contains(out, "must not say what you are about to do or just did") {
t.Errorf("prohibition must scope to process, not completed outcomes\n--- output ---\n%s", out)
}
})
}
}
// TestBuildChatPromptTwoLayerChannelPolicy pins the two INDEPENDENT axes of the
// chat channel policy (MUL-4899). Collapsing them into one condition is exactly
// the bug this matrix exists to catch:
//
// - delivery: `attachment upload` guidance is injected iff there is NO channel.
// Any IM reply leaves Multica, where the upload has nothing to bind to.
// - history: the `chat history` / `chat thread` commands are injected iff the
// channel is Slack. Those endpoints are hardwired to h.SlackHistory
// (handler/chat_history.go) — on Feishu they answer "no channel
// integration", so teaching them there sends the agent down a dead path.
//
// Feishu is the case that proves the axes are separate: no upload AND no
// history. A single `ChatChannelType != ""` gate cannot express it.
func TestBuildChatPromptTwoLayerChannelPolicy(t *testing.T) {
// Match the IMPERATIVE, not the bare command name. An IM prompt names
// `multica attachment upload` on purpose — to state that it does not apply
// here. That negation is the useful copy (the agent knows the command exists
// from the brief's Available Commands; silence would leave it guessing), so
// asserting on the bare name would forbid the very sentence we want.
const uploadGuidance = "run `multica attachment upload <local-path>`"
const historyGuidance = "multica chat history"
cases := []struct {
name string
channelType string
wantUpload bool
wantHistory bool
wantPhrases []string
}{
{
name: "direct chat: upload, no history",
channelType: "",
wantUpload: true,
wantHistory: false,
},
{
name: "slack: no upload, has history",
channelType: execenv.ChannelTypeSlack,
wantUpload: false,
wantHistory: true,
wantPhrases: []string{"Slack", "delivered to Slack as text", "You cannot attach a file to it"},
},
{
name: "feishu: no upload, no history",
channelType: execenv.ChannelTypeFeishu,
wantUpload: false,
wantHistory: false,
wantPhrases: []string{
"Feishu/Lark",
"no history reader for Feishu/Lark",
"delivered to Feishu/Lark as text",
"You cannot attach a file to it",
},
},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
out := buildChatPrompt(Task{
ChatSessionID: "sess-1",
ChatChannelType: tc.channelType,
ChatMessage: "hi",
})
if got := strings.Contains(out, uploadGuidance); got != tc.wantUpload {
t.Errorf("upload guidance present=%v, want %v\n--- output ---\n%s", got, tc.wantUpload, out)
}
if got := strings.Contains(out, historyGuidance); got != tc.wantHistory {
t.Errorf("history guidance present=%v, want %v\n--- output ---\n%s", got, tc.wantHistory, out)
}
for _, phrase := range tc.wantPhrases {
if !strings.Contains(out, phrase) {
t.Errorf("missing %q\n--- output ---\n%s", phrase, out)
}
}
})
}
}
// ChatInThread only ever selects between `chat history` and `chat thread`. With
// no Feishu history reader there is nothing to select between, so the flag must
// not leak either command into a Feishu prompt even if the server sets it.
func TestBuildChatPromptFeishuIgnoresChatInThread(t *testing.T) {
out := buildChatPrompt(Task{
ChatSessionID: "sess-1",
ChatChannelType: execenv.ChannelTypeFeishu,
ChatInThread: true,
ChatMessage: "hi",
})
for _, unwanted := range []string{"multica chat thread", "multica chat history"} {
if strings.Contains(out, unwanted) {
t.Errorf("feishu prompt must not teach %q (no Feishu history reader exists)\n--- output ---\n%s", unwanted, out)
}
}
}
func TestBuildChatPromptAgentIntro(t *testing.T) {
// The proactive self-introduction chat (MUL-4230) has no user message: the
// prompt must tell the agent to open the conversation itself, and must NOT
// carry the generic "respond to their message" framing or an empty
// "User message:" section that would confuse the agent.
out := buildChatPrompt(Task{ChatSessionID: "sess-1", ChatIntro: true})
for _, want := range []string{
"You were just created",
"you are opening the conversation",
"introduce yourself",
} {
if !strings.Contains(out, want) {
t.Fatalf("intro prompt missing %q\n--- output ---\n%s", want, out)
}
}
for _, unwanted := range []string{"Respond to their message", "User message:"} {
if strings.Contains(out, unwanted) {
t.Fatalf("intro prompt should not contain %q\n--- output ---\n%s", unwanted, out)
}
}
}
func TestBuildChatPromptSlashSkills(t *testing.T) {
t.Run("injects selected skills block", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "please [/deploy](slash://skill/abc-123) this",
Agent: &AgentData{
Skills: []SkillData{{ID: "abc-123", Name: "deploy"}},
},
}
out := buildChatPrompt(task)
if !strings.Contains(out, "Explicitly selected skills:\n- deploy\n") {
t.Fatalf("expected selected skills block, got:\n%s", out)
}
if !strings.Contains(out, "User message:\nplease [/deploy](slash://skill/abc-123) this") {
t.Fatalf("expected raw user message preserved, got:\n%s", out)
}
})
t.Run("ignores skills not belonging to agent", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "[/hacker-skill](slash://skill/evil-id)",
Agent: &AgentData{
Skills: []SkillData{{ID: "good-id", Name: "deploy"}},
},
}
out := buildChatPrompt(task)
if strings.Contains(out, "Explicitly selected skills") {
t.Fatalf("should not inject block for unknown skill ID, got:\n%s", out)
}
})
t.Run("validates by ID not label", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "[/deploy](slash://skill/wrong-id)",
Agent: &AgentData{
Skills: []SkillData{{ID: "real-id", Name: "deploy"}},
},
}
out := buildChatPrompt(task)
if strings.Contains(out, "Explicitly selected skills") {
t.Fatalf("matching label with wrong ID must not pass, got:\n%s", out)
}
})
t.Run("uses canonical name not label", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "[/spoofed-name](slash://skill/real-id)",
Agent: &AgentData{
Skills: []SkillData{{ID: "real-id", Name: "deploy"}},
},
}
out := buildChatPrompt(task)
if !strings.Contains(out, "- deploy\n") {
t.Fatalf("expected canonical name 'deploy', got:\n%s", out)
}
if strings.Contains(out, "- spoofed-name\n") {
t.Fatalf("selected skills block must not use spoofed label, got:\n%s", out)
}
if !strings.Contains(out, "User message:\n[/spoofed-name](slash://skill/real-id)") {
t.Fatalf("expected raw user message with spoofed label preserved, got:\n%s", out)
}
})
t.Run("deduplicates skills", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "[/deploy](slash://skill/a) and [/deploy](slash://skill/a) again",
Agent: &AgentData{
Skills: []SkillData{{ID: "a", Name: "deploy"}},
},
}
out := buildChatPrompt(task)
if strings.Count(out, "- deploy") != 1 {
t.Fatalf("expected exactly 1 '- deploy', got:\n%s", out)
}
})
t.Run("omits block when no valid skills", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "just a normal message",
Agent: &AgentData{Skills: []SkillData{{ID: "a", Name: "deploy"}}},
}
out := buildChatPrompt(task)
if strings.Contains(out, "Explicitly selected skills") {
t.Fatalf("should not inject block when no slash links, got:\n%s", out)
}
})
t.Run("omits block when agent has no skills", func(t *testing.T) {
task := Task{
ChatSessionID: "sess-1",
ChatMessage: "[/deploy](slash://skill/abc-123)",
Agent: &AgentData{},
}
out := buildChatPrompt(task)
if strings.Contains(out, "Explicitly selected skills") {
t.Fatalf("should not inject block for agent with no skills, got:\n%s", out)
}
})
}
// TestBuildPromptDefaultScansRootsFirst pins that the catch-all fallback
// prompt (no trigger comment, no chat, no autopilot, no quick-create)
// starts assignment-triggered comment catch-up with a bounded roots scan and
// only then offers the full-thread read, while still keeping older history
// available through pagination.
func TestBuildPromptDefaultScansRootsFirst(t *testing.T) {
out := BuildPrompt(Task{IssueID: "issue-default-1"}, "claude")
for _, s := range []string{
"multica issue comment list issue-default-1 --roots-only --summary --output json",
"--since",
} {
if !strings.Contains(out, s) {
t.Errorf("default BuildPrompt missing %q\n--- output ---\n%s", s, out)
}
}
// MUL-5372: the per-turn prompt names only the reads it wants run. Flag
// mechanics — cursors, the --recent saturation trap — live once in the
// runtime workflow file's `## Available Commands`, so restating them here
// would put the same reference text on every turn.
if strings.Contains(out, "--recent") {
t.Errorf("default BuildPrompt should not restate the --recent surface\n--- output ---\n%s", out)
}
if strings.Contains(out, "Next thread cursor:") {
t.Errorf("default BuildPrompt should not restate pagination mechanics\n--- output ---\n%s", out)
}
// MUL-5372: this path now leads with a cheap roots scan, and the scan is
// what supplies thread ids, so a generic `--thread <thread-id>` drill-down
// is well-founded here. What must still never appear is a CONCRETE anchor —
// the default path has no trigger comment to derive one from, and an
// interpolated id would send the agent after a thread that does not exist.
for _, seg := range strings.Split(out, "--thread")[1:] {
if !strings.HasPrefix(seg, " <thread-id>") {
t.Errorf("default BuildPrompt must only use the generic --thread <thread-id> placeholder, never a concrete anchor\n--- output ---\n%s", out)
}
}
// The legacy "If you need comment history" soft phrasing conflicts with
// the assignment-trigger runtime workflow, which treats reading comments
// as mandatory. Guard against it sneaking back in.
if strings.Contains(out, "If you need comment history") {
t.Errorf("default BuildPrompt still carries the legacy 'If you need' soft phrasing that conflicts with the mandatory workflow\n--- output ---\n%s", out)
}
if strings.Contains(out, "multica issue comment list issue-default-1 --output json") {
t.Errorf("default BuildPrompt still presents the unbounded flat read as the assignment catch-up command\n--- output ---\n%s", out)
}
}
// TestBuildPromptNonSquadLeaderNoRule verifies that non-squad-leader agents
// do NOT get the squad leader no_action rule injected.
func TestBuildPromptNonSquadLeaderNoRule(t *testing.T) {
task := Task{
IssueID: "issue-123",
TriggerCommentID: "comment-456",
TriggerCommentContent: "LGTM",
TriggerAuthorType: "member",
TriggerAuthorName: "Bohan",
Agent: &AgentData{
Instructions: "Some instructions without the squad marker",
},
}
out := BuildPrompt(task, "claude")
if strings.Contains(out, "Squad leader no_action rule") {
t.Errorf("buildCommentPrompt must NOT inject squad leader no_action rule for non-squad-leader agents, got:\n%s", out)
}
}
// TestBuildPromptNewCommentsHint pins that a comment-triggered task whose agent
// ran before on this issue (NewCommentsSince set, NewCommentCount > 0) gets the
// since-delta hint with the ISSUE-WIDE new-comment count, but is steered to read
// the triggering (parent) thread first rather than blindly pulling every new
// comment.
func TestBuildPromptNewCommentsHint(t *testing.T) {
const (
issueID = "issue-new-1"
since = "2026-05-28T11:00:00Z"
)
task := Task{
IssueID: issueID,
TriggerCommentID: "trigger-1",
TriggerThreadID: "thread-root-1",
TriggerCommentContent: "please look",
TriggerAuthorType: "member",
NewCommentCount: 3,
NewCommentsSince: since,
}
out := BuildPrompt(task, "claude")
// Issue-wide count (reverted from the thread-scoped wording).
if !strings.Contains(out, "3 new comment(s) on this issue since your last run") {
t.Errorf("hint must report the issue-wide new-comment count, got:\n%s", out)
}
// Don't-blindly-read-all guidance.
if !strings.Contains(out, "blindly") {
t.Errorf("hint must discourage blindly reading every new comment, got:\n%s", out)
}
// Parent thread first: the --thread <trigger> read is the prioritized action.
if !strings.Contains(out, "multica issue comment list "+issueID+" --thread thread-root-1 --since "+since+" --output json") {
t.Errorf("hint must point at the triggering (parent) thread --since read first, got:\n%s", out)
}
if !strings.Contains(out, "--tail 30") {
t.Errorf("hint must offer the full-thread (--tail 30) option, got:\n%s", out)
}
// Issue-wide catch-up is demoted to an only-if-needed fallback.
if !strings.Contains(out, "multica issue comment list "+issueID+" --since "+since+" --output json") {
t.Errorf("hint must keep the issue-wide --since catch-up as a fallback, got:\n%s", out)
}
// The old cursor-heavy paragraph must be gone.
if strings.Contains(out, "Next reply cursor") || strings.Contains(out, "--before-id") {
t.Errorf("the old cursor-pagination paragraph must not render, got:\n%s", out)
}
}
// TestBuildPromptColdStartThreadRead pins the cold-start case: no prior run means
// no since anchor (NewCommentsSince empty), so we suppress the delta hint and
// instead point the agent at the triggering CONVERSATION (--thread <trigger>
// --tail 30) rather than dumping the flat timeline.
func TestBuildPromptColdStartThreadRead(t *testing.T) {
const issueID = "issue-cold-1"
task := Task{
IssueID: issueID,
TriggerCommentID: "trigger-1",
TriggerThreadID: "thread-root-1",
TriggerCommentContent: "hi",
TriggerAuthorType: "member",
NewCommentCount: 0,
NewCommentsSince: "",
}
out := BuildPrompt(task, "claude")
if strings.Contains(out, "new comment(s) since your last run") {
t.Errorf("no since-delta hint should render on cold start, got:\n%s", out)
}
if !strings.Contains(out, "multica issue comment list "+issueID+" --thread thread-root-1 --tail 30 --output json") {
t.Errorf("cold start must point at the triggering thread read, got:\n%s", out)
}
// MUL-5372: cross-thread background is a cheap roots scan. The hint names
// only the reads it wants run — `--recent` and its saturation trap are
// documented once in the brief's `## Available Commands`, so restating the
// flag surface here would put reference text on every cold turn.
if !strings.Contains(out, "multica issue comment list "+issueID+" --roots-only --summary --output json") {
t.Errorf("cold start should offer the cheap roots scan for cross-thread background, got:\n%s", out)
}
if strings.Contains(out, "--recent") {
t.Errorf("cold start hint should not restate the --recent surface, got:\n%s", out)
}
}
// TestBuildPromptResumedNoDeltaDoesNotForceThreadRead pins the warm/no-delta
// path: when a prior provider session is actually being resumed, the triggering
// comment is already embedded in the per-turn prompt, so the agent should not
// be told to re-read the triggering thread's latest 30 replies by default.
func TestBuildPromptResumedNoDeltaDoesNotForceThreadRead(t *testing.T) {
const issueID = "issue-resumed-1"
task := Task{
IssueID: issueID,
TriggerCommentID: "trigger-1",
TriggerThreadID: "thread-root-1",
TriggerCommentContent: "hi again",
TriggerAuthorType: "member",
PriorSessionID: "session-123",
NewCommentCount: 0,
NewCommentsSince: "",
}
out := BuildPrompt(task, "claude")
for _, want := range []string{
"triggering comment is already included above",
"No other new comments on this issue since your last run",
"active thread anchor `thread-root-1` and triggering comment ID `trigger-1`",
"If your reply depends on thread context",
"do not rely only on resumed session memory",
"multica issue comment list " + issueID + " --thread thread-root-1 --tail 30 --output json",
} {
if !strings.Contains(out, want) {
t.Errorf("resumed/no-delta prompt missing %q\n--- output ---\n%s", want, out)
}
}
// The stale thread-scoped wording (since-delta used to be thread-scoped)
// must not reappear.
if strings.Contains(out, "scoped to the triggering thread") {
t.Errorf("resumed/no-delta prompt must not claim the delta is thread-scoped, got:\n%s", out)
}
if strings.Contains(out, "Read the triggering conversation first") {
t.Errorf("resumed/no-delta prompt must not use the cold-start forced-read wording, got:\n%s", out)
}
}
// TestBuildCommentPromptCoalescedCrossThread pins MUL-4195 review should-fix #3:
// when a run coalesces comments that span MULTIPLE threads, the prompt must
// embed each folded comment's content with its OWN thread id instead of
// claiming they all live in the triggering thread. The earlier version told the
// agent "they are in the triggering thread" and handed a single `--thread`
// command — wrong (and lossy) when the folded comments came from different
// threads.
func TestBuildCommentPromptCoalescedCrossThread(t *testing.T) {
task := Task{
IssueID: "issue-xthread-1",
TriggerCommentID: "trigger-newest",
TriggerThreadID: "thread-root-A",
TriggerCommentContent: "latest instruction",
TriggerAuthorType: "member",
CoalescedCommentIDs: []string{"c-old-1", "c-old-2"},
CoalescedComments: []CoalescedCommentData{
{ID: "c-old-1", ThreadID: "thread-root-A", AuthorType: "member", AuthorName: "Alice", Content: "first earlier comment", CreatedAt: "2026-07-08T01:00:00Z"},
{ID: "c-old-2", ThreadID: "thread-root-B", AuthorType: "member", AuthorName: "Bob", Content: "comment in a different thread", CreatedAt: "2026-07-08T02:00:00Z"},
},
}
out := BuildPrompt(task, "claude")
// The stale same-thread assumption must be gone.
if strings.Contains(out, "they are in the triggering thread") {
t.Errorf("prompt must not assume coalesced comments share the triggering thread, got:\n%s", out)
}
// Each folded comment's content is embedded directly, so the agent never
// has to guess which thread to read to find it.
for _, want := range []string{"first earlier comment", "comment in a different thread"} {
if !strings.Contains(out, want) {
t.Errorf("prompt must embed coalesced comment content %q, got:\n%s", want, out)
}
}
// Each distinct thread id is surfaced so a follow-up fetch targets the
// right thread — including the OTHER thread (B), not just the trigger's.
for _, want := range []string{"thread-root-A", "thread-root-B"} {
if !strings.Contains(out, want) {
t.Errorf("prompt must surface coalesced comment thread id %q, got:\n%s", want, out)
}
}
// Both coalesced comment ids remain referenced.
for _, id := range []string{"c-old-1", "c-old-2"} {
if !strings.Contains(out, id) {
t.Errorf("prompt must reference coalesced comment id %s, got:\n%s", id, out)
}
}
}
// TestBuildCommentPromptCoalescedIDsOnlyFallback pins the old-server fallback:
// when only coalesced ids are shipped (no embedded detail), the prompt must
// still NOT assume a shared thread and must point at an issue-wide fetch.
func TestBuildCommentPromptCoalescedIDsOnlyFallback(t *testing.T) {
task := Task{
IssueID: "issue-fallback-1",
TriggerCommentID: "trigger-newest",
TriggerThreadID: "thread-root-A",
TriggerCommentContent: "latest instruction",
TriggerAuthorType: "member",
CoalescedCommentIDs: []string{"c-old-1", "c-old-2"},
}
out := BuildPrompt(task, "claude")
if strings.Contains(out, "they are in the triggering thread") {
t.Errorf("id-only fallback must not assume a shared thread, got:\n%s", out)
}
if !strings.Contains(out, "--recent 30") {
t.Errorf("id-only fallback must point at an issue-wide fetch (--recent 30), got:\n%s", out)
}
for _, id := range []string{"c-old-1", "c-old-2"} {
if !strings.Contains(out, id) {
t.Errorf("id-only fallback must reference coalesced comment id %s, got:\n%s", id, out)
}
}
}
// TestCommentReplyThreadsGrouping pins the server-side grouping that drives
// per-thread reply routing (MUL-4348). The invariants:
// - three distinct root threads → three targets, each replying to its own
// thread (the trigger's thread replies under the trigger comment itself).
// - multiple coalesced follow-ups in the SAME thread → a single group, so the
// caller keeps the single-parent path and the reply is never duplicated.
// - no coalesced comments (ordinary single comment) → nil.
func TestCommentReplyThreadsGrouping(t *testing.T) {
t.Run("three distinct root threads fan out", func(t *testing.T) {
task := Task{
TriggerCommentID: "c3",
TriggerThreadID: "c3", // a root comment is its own thread
CoalescedComments: []CoalescedCommentData{
{ID: "c1", ThreadID: "c1", Content: "背一首宋词"},
{ID: "c2", ThreadID: "c2", Content: "毛泽东诗词背一首"},
},
}
targets := commentReplyThreads(task)
if len(targets) != 3 {
t.Fatalf("want 3 targets, got %d: %+v", len(targets), targets)
}
wantParent := map[string]string{"c1": "c1", "c2": "c2", "c3": "c3"}
for _, tgt := range targets {
if wantParent[tgt.ThreadID] != tgt.ParentID {
t.Errorf("thread %s: parent = %s, want %s", tgt.ThreadID, tgt.ParentID, wantParent[tgt.ThreadID])
}
}
})
t.Run("same-thread follow-ups consolidate to a single group", func(t *testing.T) {
task := Task{
TriggerCommentID: "c3",
TriggerThreadID: "thread-A",
CoalescedComments: []CoalescedCommentData{
{ID: "c1", ThreadID: "thread-A", Content: "追问 1"},
{ID: "c2", ThreadID: "thread-A", Content: "追问 2"},
},
}
if targets := commentReplyThreads(task); targets != nil {
t.Fatalf("same-thread follow-ups must not fan out; got %d targets: %+v", len(targets), targets)
}
})
t.Run("mixed: trigger thread plus one other thread", func(t *testing.T) {
task := Task{
TriggerCommentID: "c3",
TriggerThreadID: "thread-A",
CoalescedComments: []CoalescedCommentData{
{ID: "c1", ThreadID: "thread-A", Content: "same-thread follow-up"},
{ID: "c2", ThreadID: "thread-B", Content: "other thread"},
},
}
targets := commentReplyThreads(task)
if len(targets) != 2 {
t.Fatalf("want 2 targets (thread-A, thread-B), got %d: %+v", len(targets), targets)
}
got := map[string]string{}
for _, tgt := range targets {
got[tgt.ThreadID] = tgt.ParentID
}
// The trigger's own thread replies under the trigger comment, not its root.
if got["thread-A"] != "c3" {
t.Errorf("trigger thread parent = %q, want c3 (the trigger comment)", got["thread-A"])
}
// The other thread replies under the specific comment that mentioned the
// agent (a mid-thread reply), not the thread root — fixes the placement
// asymmetry from the first cut.
if got["thread-B"] != "c2" {
t.Errorf("other thread parent = %q, want c2 (the specific mentioning comment)", got["thread-B"])
}
})
t.Run("no coalesced comments → nil", func(t *testing.T) {
task := Task{TriggerCommentID: "c1", TriggerThreadID: "thread-A"}
if targets := commentReplyThreads(task); targets != nil {
t.Fatalf("ordinary single-comment run must not fan out; got %+v", targets)
}
})
t.Run("non-trigger thread replies under its newest mention, not root", func(t *testing.T) {
// Two mid-thread mentions in thread-B (oldest c1, newer c2); the reply
// should target the newest specific comment (c2), not the root thread-B.
task := Task{
TriggerCommentID: "c9",
TriggerThreadID: "thread-A",
CoalescedComments: []CoalescedCommentData{
{ID: "c1", ThreadID: "thread-B", Content: "older mention", CreatedAt: "2026-07-10T01:00:00Z"},
{ID: "c2", ThreadID: "thread-B", Content: "newer mention", CreatedAt: "2026-07-10T02:00:00Z"},
},
}
targets := commentReplyThreads(task)
got := map[string]string{}
for _, tgt := range targets {
got[tgt.ThreadID] = tgt.ParentID
}
if got["thread-B"] != "c2" {
t.Errorf("thread-B parent = %q, want newest mention c2 (not root)", got["thread-B"])
}
if got["thread-A"] != "c9" {
t.Errorf("trigger thread parent = %q, want trigger c9", got["thread-A"])
}
})
}
// TestBuildCommentPromptCrossThreadFansOutReplies is the end-to-end prompt
// assertion for the screenshot scenario: three separate root comments coalesced
// into one run must produce a per-thread reply plan (one reply per thread),
// explicitly overriding the "one comment per run" rule, instead of the single
// --parent cookbook.
func TestBuildCommentPromptCrossThreadFansOutReplies(t *testing.T) {
task := Task{
IssueID: "issue-xthread-2",
TriggerCommentID: "c3",
TriggerThreadID: "c3",
TriggerCommentContent: "莎士比亚名言来一句",
TriggerAuthorType: "member",
CoalescedCommentIDs: []string{"c1", "c2"},
CoalescedComments: []CoalescedCommentData{
{ID: "c1", ThreadID: "c1", AuthorType: "member", AuthorName: "Yushen", Content: "背一首宋词", CreatedAt: "2026-07-10T01:00:00Z"},
{ID: "c2", ThreadID: "c2", AuthorType: "member", AuthorName: "Yushen", Content: "毛泽东诗词背一首", CreatedAt: "2026-07-10T02:00:00Z"},
},
}
out := BuildPrompt(task, "claude")
for _, want := range []string{
"3 DISTINCT threads",
"Post ONE reply per thread",
"OVERRIDES",
"--parent c1",
"--parent c2",
"--parent c3",
} {
if !strings.Contains(out, want) {
t.Errorf("cross-thread prompt must contain %q, got:\n%s", want, out)
}
}
// The single-parent cookbook must NOT be used when fanning out.
if strings.Contains(out, "always use the trigger comment ID below") {
t.Errorf("cross-thread prompt must not emit the single-parent reply cookbook, got:\n%s", out)
}
// Chronological ordering (MUL-4348 test-round-2 problem #1): replies must be
// posted oldest thread first, the newest (triggering) thread last — so the
// coalesced comments c1 (oldest) and c2 come before the trigger c3.
if !strings.Contains(out, "OLDEST thread first") {
t.Errorf("cross-thread prompt must instruct oldest-first chronological order, got:\n%s", out)
}
posC1 := strings.Index(out, "--parent c1")
posC2 := strings.Index(out, "--parent c2")
posC3 := strings.Index(out, "--parent c3")
if !(posC1 >= 0 && posC1 < posC2 && posC2 < posC3) {
t.Errorf("reply targets must be listed oldest-first (c1 < c2 < c3); got positions c1=%d c2=%d c3=%d\n%s", posC1, posC2, posC3, out)
}
}
// TestBuildCommentPromptSameThreadKeepsSingleReply pins the hard requirement:
// multiple @mentions coalesced from the SAME thread must keep the ordinary
// single-parent reply path (one reply, under the trigger comment) and must NOT
// trigger the multi-thread fan-out.
func TestBuildCommentPromptSameThreadKeepsSingleReply(t *testing.T) {
task := Task{
IssueID: "issue-samethread-1",
TriggerCommentID: "c3",
TriggerThreadID: "thread-A",
TriggerCommentContent: "追问 3",
TriggerAuthorType: "member",
CoalescedCommentIDs: []string{"c1", "c2"},
CoalescedComments: []CoalescedCommentData{
{ID: "c1", ThreadID: "thread-A", AuthorType: "member", AuthorName: "Yushen", Content: "追问 1", CreatedAt: "2026-07-10T01:00:00Z"},
{ID: "c2", ThreadID: "thread-A", AuthorType: "member", AuthorName: "Yushen", Content: "追问 2", CreatedAt: "2026-07-10T02:00:00Z"},
},
}
out := BuildPrompt(task, "claude")
if strings.Contains(out, "DISTINCT threads") {
t.Errorf("same-thread coalescing must not emit the multi-thread fan-out block, got:\n%s", out)
}
// The single-parent cookbook is used, threading the one reply under the
// trigger comment.
if !strings.Contains(out, "--parent c3 --content-file ./reply.md") {
t.Errorf("same-thread run must keep the single --parent=trigger reply cookbook, got:\n%s", out)
}
}
// TestPerTurnContextBlocksCarryMovedBriefSections is the other half of
// MUL-5377: the per-run context that was removed from the runtime brief must
// still reach the agent, now via the per-turn user message. Losing it silently
// would be a worse regression than the cache cost it fixes.
func TestPerTurnContextBlocksCarryMovedBriefSections(t *testing.T) {
t.Parallel()
task := Task{
IssueID: "issue-1",
TriggerCommentID: "comment-1",
TriggerCommentContent: "please look at this",
PriorSessionResumeUnavailable: true,
InitiatorType: "member",
InitiatorName: "Bohan",
InitiatorEmail: "bohan@example.com",
ConnectedApps: []ConnectedAppData{{
Provider: "composio",
ServerName: "composio",
ToolkitSlug: "notion",
ToolkitName: "Notion",
}},
}
prompt := BuildPrompt(task, "claude")
for _, want := range []string{
"## Session Continuity Notice",
"could NOT be restored",
"## Task Initiator",
"initiated by **Bohan** (bohan@example.com), a member of this workspace",
"credentials stay scoped to the runtime owner",
"## Connected Apps",
"- Notion (`notion`) via MCP server `composio`",
} {
if !strings.Contains(prompt, want) {
t.Errorf("per-turn prompt lost moved brief content %q\n---\n%s", want, prompt)
}
}
}
// The blocks are per-run, so they must be absent when their preconditions are.
func TestPerTurnContextBlocksOmittedWhenEmpty(t *testing.T) {
t.Parallel()
prompt := BuildPrompt(Task{IssueID: "issue-1"}, "claude")
for _, banned := range []string{
"## Session Continuity Notice",
"## Task Initiator",
"## Connected Apps",
} {
if strings.Contains(prompt, banned) {
t.Errorf("per-turn prompt must not emit %q with no data\n---\n%s", banned, prompt)
}
}
}
// An assignment-triggered run carries the initiator too — it is not a
// comment-path-only block.
func TestPerTurnContextBlocksOnAssignmentPath(t *testing.T) {
t.Parallel()
prompt := BuildPrompt(Task{
IssueID: "issue-1",
InitiatorType: "agent",
InitiatorName: "GPT-Boy",
}, "claude")
if !strings.Contains(prompt, "initiated by **GPT-Boy**, another agent in this workspace") {
t.Errorf("assignment-triggered prompt lost the initiator block\n---\n%s", prompt)
}
}
// TestTurnModeMarkerAlwaysPresent is the regression guard for the review
// finding on #6021: the brief's mode router keys off an explicit marker in the
// per-turn prompt, so that marker must be emitted unconditionally from the same
// branch that selects the code path.
//
// The dangerous case is a comment-triggered run whose comment body is empty (or
// an older server that doesn't send one). Before this guard the prompt emitted
// no `[NEW COMMENT]` block at all, the brief fell through to Ownership mode,
// and the agent would change the issue status on a turn that must not.
func TestTurnModeMarkerAlwaysPresent(t *testing.T) {
t.Parallel()
cases := []struct {
name string
task Task
want string
deny string
}{
{
name: "comment-triggered with content",
task: Task{IssueID: "issue-1", TriggerCommentID: "c-1", TriggerCommentContent: "please look"},
want: "**Turn mode: Reply.**",
deny: "**Turn mode: Ownership.**",
},
{
name: "comment-triggered with EMPTY content",
task: Task{IssueID: "issue-1", TriggerCommentID: "c-1"},
want: "**Turn mode: Reply.**",
deny: "**Turn mode: Ownership.**",
},
{
name: "assignment-triggered",
task: Task{IssueID: "issue-1"},
want: "**Turn mode: Ownership.**",
deny: "**Turn mode: Reply.**",
},
{
name: "assignment-triggered with handoff note",
task: Task{IssueID: "issue-1", HandoffNote: "start with the API"},
want: "**Turn mode: Ownership.**",
deny: "**Turn mode: Reply.**",
},
}
for _, tc := range cases {
tc := tc
t.Run(tc.name, func(t *testing.T) {
t.Parallel()
prompt := BuildPrompt(tc.task, "claude")
if !strings.Contains(prompt, tc.want) {
t.Errorf("prompt missing turn-mode marker %q\n---\n%s", tc.want, prompt)
}
if strings.Contains(prompt, tc.deny) {
t.Errorf("prompt carries the wrong turn-mode marker %q\n---\n%s", tc.deny, prompt)
}
})
}
}
// The mode marker only makes sense for the two issue paths — the issue-less
// kinds have no Reply/Ownership distinction and no issue status to protect.
func TestTurnModeMarkerAbsentOnIssuelessKinds(t *testing.T) {
t.Parallel()
for _, tc := range []struct {
name string
task Task
}{
{"chat", Task{ChatSessionID: "chat-1"}},
{"quick-create", Task{QuickCreatePrompt: "make an issue"}},
{"autopilot", Task{AutopilotRunID: "run-1"}},
} {
tc := tc
t.Run(tc.name, func(t *testing.T) {
t.Parallel()
prompt := BuildPrompt(tc.task, "claude")
for _, banned := range []string{"**Turn mode: Reply.**", "**Turn mode: Ownership.**"} {
if strings.Contains(prompt, banned) {
t.Errorf("%s prompt must not carry %q\n---\n%s", tc.name, banned, prompt)
}
}
})
}
}
// The brief's router must describe the markers the prompt actually emits.
// A drift here is exactly the bug this pair of changes fixes, and it is
// invisible at runtime until an agent silently picks the wrong mode.
func TestBriefModeRouterMatchesPromptMarkers(t *testing.T) {
t.Parallel()
brief, err := execenv.InjectRuntimeConfig(t.TempDir(), "claude", execenv.TaskContextForEnv{IssueID: "issue-1"})
if err != nil {
t.Fatalf("InjectRuntimeConfig: %v", err)
}
for _, want := range []string{"`Turn mode: Reply.`", "`Turn mode: Ownership.`"} {
if !strings.Contains(brief, want) {
t.Errorf("brief mode router does not name %s\n---\n%s", want, brief)
}
}
// The retired wording keyed off the prompt's first line, which was never
// actually the [NEW COMMENT] block.
if strings.Contains(brief, "It opens with a `[NEW COMMENT]` block") {
t.Error("brief still routes on the prompt's opening line; it must route on the explicit marker")
}
}