Build the skills governance suite + the compiled knowledge base (P-SKILL.4/.5, P-SKILLREG.1/.2, P-KB.1/.2/.2b) - #242
Merged
Merged
Conversation
Seven verified increments across two epics, each fail-closed + scan-gated. Skills governance: - P-SKILL.4 (ADR-0097): the Agent Skill directory + per-skill management (inspect / re-scan / remove), closed-set trust labels, enable/disable. - P-SKILLREG.1 (ADR-0098): the enterprise registry READER seam (Ed25519 verify -> fail-closed scan-gate -> confined install). - P-SKILL.5 (ADR-0101): Skill Studio (analyze recent work -> model-drafted candidates -> gated codify). - P-SKILLREG.2 (ADR-0102): the publish seam (RegistryPublisher + LocalRegistryPublisher + fail-safe PublishDispatcher). Compiled knowledge base: - P-KB.1 (ADR-0099): the compiled-page store + gated compile pipeline (harness/kb, TS+DuckDB, frozen migration 0011). - P-KB.2 (ADR-0100): the hybrid retrieval router (vector | compiled | both) + kept-in-sync + contradiction flag. - P-KB.2b: the desktop surface (kbStore + /api/kb routes + a Compiled-KB graph view reusing mountGraph in the Knowledge panel). Invariants held: fail-closed everywhere (dead scanner => quarantine), keystone #2 (derived/registry content never auto-trusted), delimited untrusted data (#5), closed trust set (#7), frozen DuckDB migration (#10), no contracts.ts change (EventNames deferred, inv #8), no Python outside scanner-sidecar (#2). Verified: root tsc green; bun test desktop + harness green (modulo the documented fs_browse / lucid_acp Windows path-sep fails); renderer bundles; demo-P-SKILL.4/.5, demo-P-SKILLREG.1/.2, demo-P-KB.1/.2/.2b all pass; BUSL-1.1 headers present.
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
…caping or encoding' Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
This was referenced Jul 6, 2026
mlcyclops
added a commit
that referenced
this pull request
Jul 6, 2026
…d .md + three robustness layers (ADR-0177), release v1.10.3 (#246) Root cause (reproduced on the shipped artifact): extraResources stripped ALL node_modules/**/*.md; #242 made dev.ts import @oh-my-pi/pi-coding-agent in-process, which imports its prompt .md files at module load - the packaged engine died at import and never bound port 5319. - packaging: drop the !node_modules/**/*.md exclusion (READMEs are cheaper than bricks) - skills_data: the omp import is type-only + LAZY inside fail-soft discoverRaw - an unloadable optional dep degrades the Skills directory, never kills the engine (verified with the bad filter deliberately restored) - main.ts: engine stdout/stderr teed to <userData>/engine.log; the failure dialog points at the log instead of asking for a terminal relaunch - packaged_boot.test.ts: CI guard that emulates the LIVE packaging exclusions with a Bun resolver plugin and boots the real dev.ts - red in <1s on the v1.10.2 state, green on the fix, so this brick class can never ship again - version bump to 1.10.3 (version.ts + package.json + about.test.ts in lockstep) Co-authored-by: mlcyclops <mlcyclops@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
mlcyclops
added a commit
that referenced
this pull request
Jul 8, 2026
…-0186-0190) (#258) * Port P-TRIV.4/P-EVAL.1/P-CHAT.A-C pure cores onto master (ADR-0186-0190) Re-integration of the genuinely-new work from the stale feat/skills-kb-desktop branch, which forked before the skills/KB suite + Trivia Wire fixes merged to master (v1.10.2-v1.10.5). The branch's skills/KB commit was a duplicate of already-shipped PR #242 and is dropped; only these five pure cores are new: - P-TRIV.4 (ADR-0186): trivia_seed.ts - AI re-seed for the Trivia Wire. - P-EVAL.1 (ADR-0187): evals.ts - model-eval metrics + latency rollup core. - P-CHAT.A (ADR-0188): answer_sections.ts - settled-turn section splitter. - P-CHAT.B (ADR-0189): answer_chips.ts - inline tool-event chip interleave. - P-CHAT.C (ADR-0190): eval_report.ts - observed-turn -> RunRecord adapter. ADRs renumbered 0177-0181 -> 0186-0190 to clear the collision with master's released 0177-0185. Cores + tests (37) + demos (5) + typecheck all green on the master base. The app.ts/dev.ts/bridge.ts/styles.css wiring + DECISIONS.md/ PROGRESS.md are the next milestone (must re-apply against master's current chat renderer + reconcile P-CHAT.A's collapsed subagent card with master's P-TASK.5). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Wire P-CHAT.A (ADR-0188): settled-turn collapsible sections Re-apply the sectionize wiring against master's current renderer: - import answer_sections + appendIntro/appendSection + section-only renderAnswerBody in app.ts - hook the 3 settle points: renderMessage (assistant only), the `done` event, and the `finally` block - .answer-sec CSS in styles.css Subagent-card reconciliation: NO change. Master's P-TASK.5 already collapses the delegation card on finish (toggle(false)) while keeping it open during the run so live subagent activity stays visible - starting it collapsed (old P-CHAT.A) would regress P-TASK.5, so master's behavior wins. Desktop typecheck + demo-P-CHAT.A + 6 unit tests green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Wire P-CHAT.B (ADR-0189): inline tool-event chips Extend the settle transform against master's current streaming closure: - import answer_chips; add ChipData type + renderProse + createChipRow - renderAnswerBody gains the marks-aware chip branch (returns whether it chipped; a no-tool turn still sectionizes via P-CHAT.A) - the streaming closure records a mark per tool call (anchored at the answer-buffer length) inside master's existing `tool` handler; on settle the chips carry the activity and the redundant live thoughts window is dropped - .answer-chip/.tchip/.tinline CSS Desktop typecheck + demo-P-CHAT.B + 15 unit tests green. Chip/drilldown styling + collapse verified in the browser preview. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Wire P-CHAT.C (ADR-0190): settled-turn "Generate engineering report" Complete the settle transform with the run-footer CTA + server route: - bridge.ts: EvalReportTurn / EvalReportResult types + evalReport() -> POST /api/eval/report - app.ts: appendRunReport (the run footer + CTA that swaps to an "Open in Reports" link), buildEvalTurn (maps the turn's marks/tokens/failures to the observed-telemetry shape), maybeAppendReport at both settle points; the block handler now records non-quarantined tool failures for the fail-rate metric - dev.ts: the /api/eval/report route - defensively coerces the payload, maps it to evals.ts's RunRecord via renderTurnEvalReport, saves a brief (kind=brief, role=evals) so it lists in Reports - .runfoot/.report-cta/.reportlink CSS Verified END-TO-END in the browser preview: POST returns a saved brief, it lists in /api/reports, reads back via the Open-in-Reports path, and the Model-Evaluation markdown metrics compute correctly (200 tok/net-line). Full typecheck (all 3 tsconfigs) + demo-P-CHAT.C + eval tests green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Wire P-TRIV.4 (ADR-0186): Trivia Wire Settings card + AI re-seed Re-apply the Recycle feature against master's current trivia code: - bridge.ts: TriviaSeedSources / TriviaSeedView + triviaReseed() -> POST /api/trivia/reseed - dev.ts: the /api/trivia/reseed route - gathers the opt-in on-device sources (session titles / KG facts+nodes / code-graph names) and hands them to seedTrivia, which scans fail-closed and generates on the user's selected model (tool-free); fail-quiet - app.ts: secTrivia() Settings card (toggle + 3 opt-in source checkboxes + Recycle + status), reseedTrivia(), the per-role pack persistence layer (storedTriviaPack/effectiveTriviaBank/apply/clear), and the settings click handlers. Both game builders (refreshTriviaGame, ensureTrivia) now route through effectiveTriviaBank = generated pack ?? seed bank (the permanent fail-closed floor) - styles.css: the card + tightened mini source checkboxes (3px 8px padding, 14px boxes, txt-2 12.5px labels - tight + readable) Full typecheck (all 3 tsconfigs) + demo-P-TRIV.4 + 9 unit tests green. Settings card + the 3 source rows + mini-checkbox styling verified live in the browser preview. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Docs: ADR-0186-0190 entries + PROGRESS for the re-integration - DECISIONS.md: append the five ADRs (renumbered 0177-0181 -> 0186-0190) with a re-integration banner; correct the P-CHAT.A record - no subagent-card collapse (master's P-TASK.5 already handles it); update the P-CHAT status lines to WIRED (C's server route verified end-to-end) - PROGRESS.md: the shipped/stubbed/next entry for the re-integration Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Default the Trivia Wire OFF (opt-in easter egg) triviaEnabled() now returns true only when the key is explicitly "1" - an unset key (fresh install / most users) is OFF. It stays a thing people discover and switch on in Settings -> Trivia Wire, rather than showing to everyone by default. All writers already use explicit "1"/"0", so the right-click hide + undo + Settings toggle are unaffected. Verified in preview: fresh origin -> wire not mounted, toggle unchecked; flipping the toggle writes "1" and mounts the wire. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: mlcyclops <mlcyclops@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Seven verified increments across two epics, rebased cleanly on top of the Trivia Wire merge (#241). Every increment is fail-closed + scan-gated; no
contracts.tschange (EventNames deferred).Skills governance
RegistryPublisher+ defaultLocalRegistryPublisher+ a fail-safePublishDispatcher(mirrors the SIEM Sink); round-trips through the P-SKILLREG.1 reader.Compiled knowledge base (OpenKB-style)
harness/kb, TS+DuckDB, frozen migration0011): scan source → compile → re-scan every derived page → store. A poisoned source or page is quarantined, never stored.vector | compiled | both, delimited + cited) + the kept-in-sync generator (idempotent, contradiction-flagged, prior page retained).kbStore()+/api/kb/ingest|retrieve|graphroutes (backed bybackend.complete) + a Compiled KB graph view in the Knowledge panel, reusingmountGraphoverkb_links(node click → page body as escaped DATA).Invariants held
Fail-closed everywhere (dead scanner ⇒ quarantine) · keystone #2 (derived/registry/drafted content never auto-trusted) · untrusted content delimited (#5) · closed trust set (#7) · frozen DuckDB migration (#10) · EventNames deferred, no
contracts.tsbytes (#8) · no Python outsidescanner-sidecar/(#2).Verification
bun x tsc --noEmit: green.bun test desktop+bun test harness: green modulo the documentedfs_browse/lucid_acpWindows path-sep failures (pre-existing, unrelated).make demo-*targets pass (real scanner + real Ed25519 where relevant); BUSL-1.1 headers present.Notes
origin/master(c7bdc8a, the Trivia Wire); my stale local trivia copies were discarded in favor of the merged final versions.registry/publishremote implementations and the RAG/vector desktop wiring remain follow-ups (seams are in place; retrievevector/hybridcurrently degrade to compiled hits until the vector store is desktop-wired).