Optimize Boring Backend skill routing and verification#3
Conversation
dd3ok
left a comment
dd3ok
left a comment
There was a problem hiding this comment.
Self-review complete.
- No reproducible P0-P4 code, routing, mirror, or documentation findings remain.
- OpenAI/Anthropic packaging and progressive-disclosure guidance checked: concise third-person discovery metadata, direct one-level references from
SKILL.md, and conditional detail loading are preserved. - Duplicate/stale-text scan found only intentional verifier fixtures; runtime package has no repeated long lines.
- Ordinary subagent guidance is isolated from experiment/handoff rules, and token reporting is no longer on the default routing path.
- Local verification passed (33 repository tests, 8 forward implementation tests, Python compilation, patch whitespace check); GitHub Ubuntu/macOS/Windows checks also passed.
Residual risk: token savings are word/byte proxies, not a guarantee of identical tokenizer counts or behavior across every Codex/Claude model. Live multi-model evaluations remain the appropriate follow-up for release-level confidence.
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request restructures the boring-backend skill by moving experiment reporting, subagent delegation, and token reporting guidelines into dedicated reference files, updating the skill mirrors accordingly. It also introduces a unified verification entrypoint script verify_all.py with corresponding unit tests and CI workflow configurations. The review feedback highlights opportunities to improve the robustness of a case-sensitive assertion in the unit tests and to simplify redundant string checks in the semantic validation script.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2c5478ce92
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| r"\bfor\b.*(?:\bcompatibility\b|backup/restore)", line | ||
| ): |
There was a problem hiding this comment.
This narrowed pattern only catches overlap wording that says for compatibility after operations-guard-catalog.md; a bad route such as “Route compatibility to references/operations-guard-catalog.md” or “use operations when compatibility changes” now passes even though the guard is meant to prevent compatibility/backup-restore work from being routed to operations. Since this verifier is the protection against catalog-routing drift, keep the new catalog-inventory line from false-positiveing without requiring this specific for ... phrasing.
Useful? React with 👍 / 👎.
1 participant
Summary
SKILL.mdfor vendor-compatible progressive disclosureToken impact
Word-count proxy against
main:SKILL.md: 589 -> 508 words (-13.8%)SKILL.md+ core routing: 951 -> 858 words (-9.8%)The experiment and token-reporting references load only when their trigger applies. Catalogs remain conditional and require a
catalog_routereason.Verification
python scripts/verify_all.py(33 repository tests and 8 forward implementation tests)python -m py_compile scripts/verify_boring_backend_skill_mirrors.py scripts/verify_all.py tests/test_verify_boring_backend_skill_mirrors.py tests/test_verify_all.pygit diff HEAD^ HEAD --check.agents, and.claudepackage hashes verified identicalReview notes
git diff --checkwas intentionally not added to CI because a clean checkout has no working-tree diff; local and pre-PR range checks cover the changed patch.