[bk-agent-01]

Closes TEC-1818

Summary

Agent-toolkit results are now short by default. Next-call hints now use the full session id, so they always resolve.

One t3_find_sessions line, before:

1. `86157111` **Compact output** · project bkt3 · TEC-1818 · TEC-1624 · PR #301 (open, checks passing) · PR #300 (merged, checks passing) · PR #299 (closed) — idle, last change 1m ago

After (default detail: "compact"):

1. `86157111` **Compact output** · project bkt3 · 2 tickets · 3 PRs (1 open) — idle, last change 1m ago

detail: "full" gives the old line exactly.

What changed:

1. t3_find_sessions, t3_inbox, t3_answer and t3_session_brief take detail: "compact" | "full". The default is compact. A single ticket or pull request still shows as itself.
2. In JSON mode, compact replaces the linear and pullRequests arrays with linearCount, pullRequestCount and openPullRequestCount. Full keeps the arrays.
3. Every generated next-call hint uses the full session id. The short ref stays for display. Subagent ids (thread:delegated-task:...) can share a short ref, so a short ref can match several sessions.
4. t3_list_sessions takes an opaque cursor and returns nextCursor. The order is the same as before (newest change first). It now drops linearLinks to linearLinkCount unless detail: "full".
5. Question options, request ids and answer calls are not changed.

Decisions

Decision	Options	Why	Undo	Confidence
t3_whoami and t3_wait also use compact tags	Compact everywhere / only the four named tools	They share the same heading code. The ticket goal is "short by default".	Pass the detail to their context	Sure
Request ids in hints stay short	Full / short	A request id resolves only inside one session, so it cannot collide with another session.	Use the full request id in requestRef	Sure
Schedule ids in hints stay short	Full / short	Out of scope: the ticket is about session ids. Session hints in schedules and watch notes now use the full id.	Change shortRef(task.id) calls	Sure
t3_list_sessions cursor	Offset / keyset on the last session / both	The cursor keeps the offset and the last session. The next page starts after that session when it did not change, so a new session does not shift the page. It falls back to the offset.	Use the offset only	Sure
t3_list_sessions compact by default	Default compact / default full	The ticket asks for shorter default output. Callers that read linearLinks must pass detail: "full". No caller in this repository reads it.	Make full the default	Sure

Acceptance check

Ticket item	Proof	Status
Compact heading is the default; full keeps today's output	compactOutput.test.ts "session heading detail"	passed in CI run 40739549318
JSON compact gives counts; full gives arrays	compactOutput.test.ts "session tags as data"	passed in CI run 40739549318
Next-call hints use the full session id	compactOutput.test.ts "next-call hints", render.test.ts, scheduledTaskWatch.expbkt3.test.ts	passed in CI run 40739549318
Question options and request ids are kept	compactOutput.test.ts "names the full session id in t3_inbox hints"	passed in CI run 40739549318
t3_list_sessions cursor pages, compact linearLinks	handlers.listSessionsPaging.test.ts	passed in CI run 40739549318

How I tested

No local tests on the shared host; CI runs the suite.

CI run 40739549318 on head 2de0a3039: Lint, Build and Typecheck, Fork markers and the new tests pass (compactOutput.test.ts 8, handlers.listSessionsPaging.test.ts 5, render.test.ts 6, scheduledTaskWatch.expbkt3.test.ts 7). Test Server 4 fails in src/storageCleanup.test.ts > "compares the checked-out branch without a full status". That failure is already on stage: the last run of PR #354 (40705333893) fails the same test. This PR does not touch storage cleanup.

Cross-review: not run yet. The orchestrating session owns the review step.

🤖 Generated with Claude Code

[bk-agent-01]

[cross-review r1 · gpt-6.1-sol] P2 Two correctness gaps remain: cursor recovery after an anchor disappears, and session references in wait continuation hints. No local tests ran. The storageCleanup failure also appears in base run 40705333893; this PR does not change that code.

[bk-agent-01]

Cross-review record — gpt-6.1-sol on a9cc2d4

Rounds: 2. Verdict: clean.

Round	Finding	Severity	Outcome (commit or reason)
1	r1-1: Cursor fallback skips sessions after anchor deletion	P2	Fixed in a9cc2d4. The cursor stores the last sort key. The next page selects keys strictly after it. Tests cover deletion, anchor movement, and equal timestamps.
1	r1-2: Wait hints retain ambiguous input references	P2	Fixed in a9cc2d4. Both affected hints use the resolved full session IDs through waitAgainHint. Tests cover later prefix ambiguity and an unrestricted wait.
2	No new findings	—	Reviewed the delta and affected callers. The caller search found no dependency on snapshot order for equal timestamps.
Deferred: none. Fallback: none.

No local tests, builds, or linters ran. The known storageCleanup test failure is outside this review's scope.

[bk-agent-01]

Merge decision: tushar chose "AI fixes it" (test fix, then merge B1 and B2 to stage) in the T3 session on 2026-10-29. Head d0a97be, all checks green. Cross-review: gpt-6.1-sol clean after 2 rounds on a9cc2d4; this head only adds a clean merge of stage (#357). bkmain promotion stays tushar's call.