Skip to content

Improve agent token efficiency for history and activity #35

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
unbraind merged 1 commit into from
May 20, 2026
Merged

Improve agent token efficiency for history and activity #35

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 19 additions & 2 deletions .agents/pm/features/pm-rnpb.toon
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,15 +6,16 @@ status: in_progress
priority: 0
tags[10]: agent-ux,audit,calendar,ci,dogfood,pm-cli,sdk,search,security,telemetry
created_at: "2026-05-03T13:20:01.343Z"
updated_at: "2026-05-20T21:10:28.981Z"
updated_at: "2026-05-20T22:23:02.164Z"
deadline: "2026-06-01T00:00:00.000Z"
assignee: codex-audit-2026-05-20
author: codex-audit
estimated_minutes: 300
acceptance_criteria: Audit findings are actionable and deduplicated; changed files and verification commands are linked; temporary dogfood project validates core workflows; any discovered bugs with safe fixes are remediated and tested.
parent: pm-jrjt
dependencies[1]{id,kind,created_at,author,source_kind}:
pm-jrjt,parent,"2026-05-03T13:20:01.343Z",codex-audit,null
comments[139]{created_at,author,text}:
comments[155]{created_at,author,text}:
"2026-05-03T13:20:01.343Z",codex-audit,Parent lineage established under pm-jrjt after duplicate checks found only closed prior audit items.
"2026-05-09T11:12:26.232Z",codex-agent,"Duplicate check before this pass: pm context --limit 10 --depth standard, pm search dogfood audit extensible sdk calendar semantic search telemetry sentry --limit 10, pm list-open --limit 20, and pm list-in-progress --limit 20. Reusing open feature pm-rnpb; prior same-day task pm-m35h is closed, so no duplicate item created."
"2026-05-09T11:46:59.430Z",codex-agent,"Evidence 2026-05-09: temp dogfood project exercised init/create/claim/update/files/docs/tests/calendar/context/aggregate/dedupe/search/reindex/extension scaffold/install/doctor/contracts/manage. Fixed search gap discovered there: reminder/event metadata is now in keyword and semantic corpus, and short phrase scoring ranks exact calendar event titles first. Verification passed: focused search/reindex/cache tests, release-readiness/search focused tests, linked pm test, full 100% coverage, static quality gate, CI workflow contract tests, pm health, pm validate, npm pack dry-run, secret scan, live telemetry/Sentry checks, and real hybrid reindex of 679 items."
Expand Down Expand Up @@ -154,6 +155,22 @@ comments[139]{created_at,author,text}:
"2026-05-20T20:44:34.646Z",claude-code-agent,"text=Dogfood+privacy session 2026-05-20 (claude-code-agent, sub-agent driven). Shipped under PR #34 (branch audit/dogfood-2026-05-20-agent-ux): pm-qrxs (MCP pm_search compact default), pm-8jd3 (MCP activity compact default), pm-h4zb (secret guardrail: all gh token prefixes + testable rules + sudo-pw redaction), pm-atsv (CLI consistency: append text forms, scope-error enumeration, docs/test --list, command did-you-mean), pm-d7us (Sentry PM-CLI-R/PM-CLI-S root-caused as pre-2026.5.18 unguarded .trim/.join, already fixed on HEAD, marked resolvedInNextRelease). Gates: build, typecheck, 100% coverage (1620/1179/279/1566), secret scan, static-quality, packaged dogfood (84 cmds), sentry/telemetry gate ok (sentry 0/0/0, telemetry error rate 2.6%, queue draining real-time). Open follow-ups: pm-3pbs, pm-nbht, pm-orrl, pm-5k2w."
"2026-05-20T21:01:07.352Z",codex-pr-review,"Evidence: addressed PR #34 review threads. Local verification passed: pnpm build; focused suite PM_RUN_TESTS_SKIP_BUILD=1 node scripts/run-tests.mjs test -- tests/unit/check-secrets.spec.ts tests/unit/get-append-command.spec.ts tests/unit/files-docs-command.spec.ts tests/integration/help-runtime.spec.ts tests/integration/mcp-dynamic-package-actions.spec.ts tests/unit/test-command.spec.ts (6 files, 104 tests); full gate pnpm build && node scripts/run-tests.mjs coverage && pnpm quality:static && node scripts/check-secrets.mjs (96 files, 1392 tests, 100 percent coverage, static gate, secret scan)."
"2026-05-20T21:09:02.005Z",codex-pr-review,"GitHub verification: PR #34 commit 03765c95 has mergeStateStatus CLEAN and all checks passed: Build foundation, Ubuntu tests, macOS tests, coverage, typecheck, static, compat, smokes, Sourcery, and CodeRabbit (Actions run 26189749692). Proceeding to merge and delete PR branch."
"2026-05-20T21:14:13.993Z",codex-audit-2026-05-20,"Duplicate-check evidence: searched manual dogfood SDK calendar search performance CI telemetry token agent UX and reused canonical feature pm-rnpb; open related follow-ups include pm-59gj, pm-gt82, pm-6vfg, pm-orrl, pm-nbht, pm-3pbs, pm-mbdu, and pm-5k2w."
"2026-05-20T21:19:00.923Z",codex-audit-2026-05-20,"Sub-agent evidence: search/calendar/vector review found reindex does not prune deleted vector records, timezone-aware calendar bucketing/display gaps, explicit calendar --format toon falling back to markdown, and extension-backed reindex underreporting batches_completed. SDK/package review found permissive runtime extension action schemas, hardcoded optional package command ownership maps, catalog privacy inference, and duplicated capability constants. GitHub/Sentry/privacy review found no open repo issues, Dependabot PRs green, Sentry unresolved 0, tracked secret scan clean, but private telemetry stack has a queue backlog requiring ops follow-up. Reusing existing pm-rnpb/pm-59gj/pm-xk39/pm-gt82 lineages rather than duplicating items."
"2026-05-20T21:31:49.894Z",codex-manual-dogfood,"Manual dogfood reviewer start: bootstrapped with npm install -g ., pm --version=2026.5.18, node=v25.9.0, pnpm=10.33.0, pnpm build passed. Reusing active audit lineage; dogfood commands will use isolated temp PM_PATH/PM_GLOBAL_PATH and no repo file edits."
"2026-05-20T21:34:12.959Z",codex-manual-dogfood,"Manual dogfood findings 2026-05-20: isolated temp project at /tmp/pm-manual-dogfood-20260520-o4D9ev with PM_PATH/PM_GLOBAL_PATH separated. Covered init --with-packages, create/claim/update/close/release, files/docs/tests/comments/notes/learnings, list/search/context/activity/history, deps lookup, calendar, package catalog/explore/manage/doctor/install, contracts/help, todos/beads import-export surfaces. Key findings: 53 MODULE_TYPELESS_PACKAGE_JSON warnings from bundled extensions after package install; pm comments on long-lived item returns full history; activity default still prints full patches; dependencies command expectation fails because actual command is deps; calendar event create phrase is silently treated as agenda; calendar same-day from/to rejected; package explore has no query/target filter; todos export silently ignores unsupported --format/--output and writes default folder; package/catalog invalid fields guidance is helpful; history --compact works and is much smaller; linked test sandboxing works with node scripts/run-tests.mjs form."
"2026-05-20T21:34:56.833Z",codex-gpt5,"Sub-agent finding (search/calendar/vector): semantic reindex can use extension embed providers but search requires full query providers; gc embeddings leaves vectorization-status/lancedb state; auto-index refresh warnings are profile-only; completions advertise optional package commands; LM Studio works through openai.base_url but is not discoverable. Implementing safe cleanup fix now; larger provider/completion UX gaps remain tracked under pm-rnpb follow-ups."
"2026-05-20T21:36:19.701Z",codex-gpt5,"Sub-agent finding (performance/CI/code quality): startup --version fast path is healthy around 40ms, but normal CLI imports still cost ~176-377ms and release-readiness/history focused specs are slow due to repeated spawned CLI calls. Static quality gate passes but duplicate/complexity focus misses CLI command hotspots. Keeping this branch to compact history + semantic gc cleanup; performance/test-sharding/static-quality ratchets remain in existing pm-gt82/pm-mbdu/pm-rnpb follow-ups."
"2026-05-20T21:44:55.222Z",codex-full-audit,"Audit kickoff 2026-05-20: reused existing parent pm-rnpb and in-progress child pm-3pbs after duplicate search. Bootstrap passed: npm install -g ., pm --version 2026.5.18, node v25.9.0, pnpm 10.33.0, pnpm build. Preserving existing dirty implementation changes before further edits."
"2026-05-20T21:48:12.269Z",codex-full-audit,"GitHub external audit 2026-05-20: no open issues, Dependabot alerts, secret scanning alerts, private security advisories, queued runs, or unmerged review work. Actionable maintenance PRs found: #30 pnpm/action-setup 6.0.4->6.0.8 and #31 tsx 4.22.0->4.22.1 are clean/green; Dependabot Updates run 26014999094 had a one-off unknown_error on softprops/action-gh-release but created PR #30 successfully. No new duplicate pm item created; tracking under pm-rnpb unless the failure repeats."
"2026-05-20T21:49:01.863Z",codex-full-audit,"CI/coverage audit 2026-05-20: current 100% gate is strict but not broad because release-readiness coverage allowlist excludes critical runtime surfaces such as cli main, telemetry/Sentry runtime, MCP server, SDK contracts, and command registration. Recent CI wall time about 7m with coverage ~5m41; high-churn cache keys hash src/tests/tracker content; docs-only PRs still run full PR CI; release workflow repeats most gates; auto-release watches active Release workflows without pinning to the dispatched tag. Track as pm-rnpb follow-up; no duplicate item created during this focused patch."
"2026-05-20T21:50:07.683Z",codex-full-audit,"Manual dogfood sub-agent 2026-05-20: verified isolated temp project with dist/cli.js 2026.5.18 and separate PM_PATH/PM_GLOBAL_PATH. Healthy paths: init, progressive create, claim/update/files/docs/tests/comments/notes/learnings/deps/test-run/context/list/get/history/activity/search/reindex/guide-shell/calendar/validate/normalize/gc/close/release. Findings mapped to existing items, no duplicates: health --brief still 369 lines -> pm-nbht; init/config/package list bootstrap output noisy -> pm-5k2w/pm-59gj; dependencies alias absent -> pm-5k2w; event same-start/end recovery needs duration example -> pm-5k2w. Current patch addresses pm-3pbs history/activity verbosity."
"2026-05-20T21:51:22.682Z",codex-full-audit,"Telemetry/Sentry audit 2026-05-20: Sentry currently has 0 unresolved issue groups and no recent error logs; recent logs are usage warnings. Live telemetry API accepts events, but ingestion is stalled: persisted data is stale since 2026-05-19T21:45:44Z and the event queue has a large ready/unacked backlog while worker logs show insert failures. pm health --check-only --json reported telemetry ok despite stale persistence/backlog, so release/health gates need freshness/backlog checks. Host-specific diagnostics remain only under ignored local telemetry tooling; no private stack details added to tracked docs."
"2026-05-20T21:52:08.303Z",codex-full-audit,"SDK/extensions audit 2026-05-20: package-first review found init --with-packages can report installed_all while package doctor shows first-party package runtime load failures for package runtimes using top-level await under the current transform; extension action schemas remain permissive/type-poor; strict extension flag validation needs required/short flag parity; SDK/contracts should expose richer config-driven schema metadata; contracts --command should accept extension command groups; package-owned command hints are still partly hardcoded. Mapped to existing pm-rnpb/pm-59gj plus current pm-3pbs strict-flag patch; no duplicate item created."
"2026-05-20T22:03:30.428Z",codex-full-audit,"Code/performance audit 2026-05-20: confirmed activity full is now exposed in action schema; remaining performance gaps are follow-ups: compact history/activity reduce tokens but still read/sort full histories, GC/search artifact registry is duplicated between gc and cache invalidation, and read command dispatch still imports a broad command index. Strict extension validation now rejects unknown/disabled long flags and canonicalizes short flags in the parser helper; full runtime short/required parity remains a pm-rnpb/pm-59gj follow-up because command dispatch does not yet consistently preserve short option tokens for extension subcommands."
"2026-05-20T22:10:49.653Z",codex-full-audit,"PR evidence 2026-05-20: pushed branch audit/token-output-activity-history-2026-05-20 commit 50d80641 and opened PR #35 (https://github.com/unbraind/pm-cli/pull/35) for compact history/activity output, semantic gc cleanup, and dynamic flag validation. Local gates passed before push: build, focused regression, full coverage 100%, static quality, secret scan. Watching CI and review-bot feedback before merge."
"2026-05-20T22:20:23.463Z",codex-full-audit,"PR #35 review feedback addressed: Sourcery noted normalizeActivityOptions ignored explicit compact=false when full was absent; Gemini noted extension subcommand flag lookup precedence. Patched activity compact precedence and made nested extension command flag definitions win over group-level matches. Focused verification passed: PM_RUN_TESTS_SKIP_BUILD=1 node scripts/run-tests.mjs test -- tests/integration/cli.integration.spec.ts tests/unit/main-loose-options.spec.ts tests/unit/history-activity-command.spec.ts tests/integration/mcp-dynamic-package-actions.spec.ts --reporter=dot (4 files, 102 tests); pnpm exec tsc --noEmit passed."
"2026-05-20T22:23:02.164Z",codex-full-audit,"PR #35 CodeRabbit follow-up addressed: preserved alias values when both canonical and alias extension flags are supplied, while keeping nested extension flag precedence. Focused verification passed again: PM_RUN_TESTS_SKIP_BUILD=1 node scripts/run-tests.mjs test -- tests/unit/main-loose-options.spec.ts tests/integration/cli.integration.spec.ts tests/unit/history-activity-command.spec.ts tests/integration/mcp-dynamic-package-actions.spec.ts --reporter=dot (4 files, 102 tests); pnpm exec tsc --noEmit passed."
notes[2]{created_at,author,text}:
"2026-05-17T07:36:59.474Z",codex-agent,"Design note 2026-05-17: unfiltered pm contracts default now intentionally omits three heavy sections (schema oneOf union, command_flags, commander_aliases) and emits explicit omitted_reason markers; --full restores complete schema and alias/flag surfaces for tooling that needs maximal detail."
"2026-05-17T08:33:25.981Z",codex-agent,"Design note 2026-05-17: default brief projection was intentionally scoped to bare 'pm list' instead of all list-* commands to preserve existing automation contracts while still cutting agent token usage on the highest-frequency exploratory list call."
Expand Down
Loading