This repository was archived by the owner on Apr 9, 2026. It is now read-only.
Open
Conversation
Platform adapter architecture + 4 platform implementations: - Codex CLI: AGENTS.md sync + ~/.codex/hooks.json (SessionStart, UserPromptSubmit) - Gemini CLI: GEMINI.md + .gemini/settings.json hooks (BeforeTool, AfterTool, PreCompress, SessionStart) + commands/compound.toml - OpenCode: OPENCODE.md + opencode.json instructions + .opencode/plugins/tenetx.ts scaffold (25+ hook events, npm-native) - Copilot CLI: .github/copilot-instructions.md + .github/hooks/tenetx.json (preToolUse deny/allow) + .github/agents/compound.agent.md + CLAUDE.md native compatibility Usage: tenetx init codex|gemini|opencode|copilot tenetx sync codex|gemini|opencode|copilot Core adapter (platform/adapter.ts): - generateSolutionInstructions(): verified+ solutions → markdown - syncToInstructionFile(): platform-specific instruction file with markers Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rewrote all platform adapters from stubs to full implementations: Gemini CLI (highest parity): 6 hook scripts: BeforeTool (danger+reflection), AfterTool (negative+write+hints), BeforeAgent (solution injection+prompt learning), SessionStart (extraction+lifecycle), PreCompress (compound hint), SessionEnd (workflow completion) Copilot CLI: 4 hook scripts: preToolUse (danger+reflection with deny/allow), postToolUse (negative+write tracking), userPromptSubmitted (prompt learning), sessionStart (extraction+lifecycle). Agent: compound.agent.md OpenCode: Full TS plugin with real logic: tool.execute.before (danger+reflection), tool.execute.after (negative+write), session.idle (sync), session.compacting (context injection). updateEvidence() included. Codex CLI (3-hook maximum): SessionStart (extraction+lifecycle+sync), UserPromptSubmit (prompt learning), Stop (workflow completion+resync). Limited by platform. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Mark non-Claude platforms as experimental with clear feature parity table. Note that current implementation uses generated shell scripts calling tenetx engine, and native platform packages are planned for v2.2.0. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Evidence counters (injected, reflected, negative, sessions) were silently failing in ESM runtime because require() is not defined in "type": "module". All try/catch blocks caught the error silently, so tests passed but evidence never incremented — making the entire lifecycle system inert. Fixed: pre-tool-use.ts updateSolutionEvidence() and post-tool-use.ts updateNegativeEvidence() now use async/await import(). Also fixed: CHANGELOG footer links for v2.0.0, v1.7.0, v1.6.3. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- checkCompoundReflection → async, await updateSolutionEvidence calls
- Call site uses .catch() for non-blocking promise handling
- solution-injector: await on updateSolutionEvidence in injection loop
- search.ts: require('node:os') → import * as os from 'node:os'
Without these fixes, evidence counters (reflected, sessions, injected)
would fire-and-forget their promises, never actually writing to files.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- post-tool-use: updateNegativeEvidence() now has .catch() to prevent unhandled promise rejection in Node.js - pack/search: packName sanitized before shell interpolation in git commit (removes non-alphanumeric chars to prevent command injection) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
wooo-jin
added a commit
that referenced
this pull request
Apr 9, 2026
The Round 3 v2 fixture exposed 4 hard-positive ranking failures where
hyphenated solution tags (api-key, code-review, red-green-refactor) only
intersected query tokens that contained literal hyphens. extractTags
strips hyphens during query extraction (api-key → [api, key]), so
compound tags only matched via the half-weight partialMatches fallback,
losing to competitors with single-word direct hits.
Fix: two pure helpers in solution-format.ts that recover compound forms
from BOTH sides of the matching pipeline:
expandCompoundTags(tags) — solution-side. 'api-key' → [api-key, api, key]
expandQueryBigrams(tags) — query-side. ['api','keys'] → +[api-key,
apikey, api-keys, apikeys]
Both helpers are layered into calculateRelevance via a new optional
solutionTagsExpanded field on CalculateRelevanceOptions. The expanded
sets are used for intersection/partialMatches, but the Jaccard union
denominator still uses the RAW solution tags so score normalization
stays semantically stable (no asymmetric inflation).
Empirical impact (fixture v2, 53+16+14 queries):
| metric | baseline | R4-T1 | delta |
|-----------------------|----------|-------|--------|
| recallAt5 | 1.000 | 1.000 | = |
| positive mrrAt5 | 0.959 | 0.981 | +0.022 |
| paraphrase mrrAt5 | 1.000 | 1.000 | = |
| aggregate mrrAt5 | 0.969 | 0.986 | +0.017 |
| negativeAnyResultRate | 0.357 | 0.357 | = |
2 of 4 v2 hard positives flipped to @1:
- "managing api keys and credentials safely" → starter-secret-management
- "red green refactor cycle for new features" → starter-tdd-red-green-refactor
The other 2 (avoiding hardcoded credentials, writing unit tests for a
function with side effects) are query-side English semantics — R4-T2
or R4-T3 territory, not R4-T1's scope.
negativeAnyResultRate unchanged: R4-T1 is a ranking-quality fix, not
an FP filter. R4-T2/R4-T3 will attack that surface.
Hook path (rankCandidates) and MCP path (searchSolutions) are updated
identically — both wire expandQueryBigrams BEFORE normalizeTerms and
expandCompoundTags per solution. ROUND3_BASELINE updated to v3 with
inline history block + per-case mechanism notes.
Test additions:
- 15 unit tests for the two helpers (Korean preservation, dedup,
Unicode safety, plural stem edge cases, ASCII filter)
- 1 integration regression guard asserting the 2 specific R4-T1 hard
positives reach @1 (not just aggregate mrrAt5 — silent flip-back
protection)
- New evaluateQuery() export for per-query test assertions
Reviewed by code-reviewer (2 MED + 2 LOW initially → MED #2 fixed in
this PR, MED #1 deferred to R4-T2 per reviewer guidance, LOWs skip)
and security-reviewer (0 findings, APPROVE).
All 1274 tests pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
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.
Summary
Commits (6)
0666604feat: multi-platform support (Codex, Gemini, OpenCode, Copilot)dbe2e08feat: full compound learning on all 4 platforms (Claude parity)70cf1bcdocs: honest multi-platform documentation28170c2fix(critical): require() → await import() for ESM evidence counterscf327a6fix(critical): async/await for evidence counters + ESM require fixf697181fix: unhandled promise rejection + shell injection sanitizationTest plan
🤖 Generated with Claude Code