Repository navigation
Add .claude/skills for PR workflow, testing tiers, and review rounds - #960
Merged
Merged
Conversation
Contributor
|
❌ Automatic Labeling Failed An error occurred while trying to automatically label this pull request. Please check the workflow logs for details and add labels manually. |
Adds .claude/skills/ for the agent development workflow in this repo: - open-pr: branch, identity, classify the change's track, PR checklist - testing: unit tier (what ci.yml runs) always, plus a real end-to-end run (zero-API local, or a dispatched Integration tests/Benchmarks CI leg) when the change can affect one - dashboard-change: build + before/after screenshots (Playwright MCP if connected, else the repo's own scripts/tools/shoot_dashboard.py) instead of a real run, for dashboard/frontend-only changes - docs-change: link check only, for docs-only changes - review-cycles: 3 rounds of two independent reviewers in parallel (GLM via Claude Code, IBM Bob Shell, the user's own Claude Code as fallback), ported from rossoctl/context-guru's offload-review skill (Apache-2.0) and adapted to this repo's house rules and a 3-round default; see scripts/offload-review/NOTICE.md for what changed CLAUDE.md is rewritten short: what the repo is, a table pointing at the skills above, and a DCO reminder (every commit needs Signed-off-by, checked by this repo's probot/dco CI check) — the detailed provider-order prose moved into the testing skill. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Signed-off-by: Osher Elhadad <Osher.Elhadad@ibm.com>
…xists core/tests ran against this PR's own new ./.claude/skills/ (the dev- workflow skills this PR adds) and failed: _find_skills_dir() picked it as soon as Path(cand).is_dir() was true, without checking it is actually a built cap-evolve skills registry. It has no _registry/manifest.json, so _resolve_skills() then raised FileNotFoundError downstream in cap-evolve run. This was already a latent bug for any host whose .claude/skills/ (or ~/.claude/skills/) holds unrelated, non-cap-evolve skills — not specific to this PR's directory. Require _registry/manifest.json to exist before accepting a candidate; add a regression test. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Signed-off-by: Osher Elhadad <Osher.Elhadad@ibm.com>
Signed-off-by: Osher Elhadad <Osher.Elhadad@ibm.com>
OsherElhadad
force-pushed
the
chore/agent-dev-skills
branch
from
November 8, 2026 22:11
051767d to
f4e0116
Compare
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
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.
What & why
Gives AI agents (and anyone following CLAUDE.md) a repeatable, repo-aware process
for development work in cap-evolve: opening a PR, picking the right testing tier,
and running real review cycles — instead of ad-hoc behavior per session.
Adds
.claude/skills/:open-pr— branch offmain, commit under the right identity, classify thechange's track (code / dashboard / docs-only), fill the PR checklist.
testing— the unit tier CI already runs, always; plus a real end-to-endrun (free zero-API locally, or a dispatched
Integration tests/Benchmarksworkflow run) whenever the change can plausibly affect one. Includes the model
provider order (RITS → ETE-INT → ETE), moved here from
CLAUDE.md.dashboard-change— fordashboard/frontendchanges: build + before/afterscreenshots (Playwright MCP if connected, else the repo's existing
scripts/tools/shoot_dashboard.py) attached to the PR, no real end-to-end runneeded.
docs-change— for docs-only changes: a locallycheelink check, thenstraight to review — no run, no screenshots.
review-cycles— 3 independent review rounds per PR, each with two parallelreviewers (GLM via Claude Code, IBM Bob Shell, or the user's own Claude Code as
fallback). Ported from
rossoctl/context-guru's
offload-reviewskill (Apache-2.0) — seescripts/offload-review/NOTICE.mdforexactly what changed (3 rounds instead of 2, cap-evolve's own house rules in the
review prompts, dropped a context-guru-specific policy).
CLAUDE.mdis rewritten short: what the repo is, a table pointing at the skillsabove, and a DCO reminder (this repo's
probot/dcocheck fails any PR with anunsigned commit — always
git commit -s). The detailed model-provider prose movedinto the
testingskill;.bob/rules/01-cap-evolve.mdnow points at the skillstoo.
Checklist
python -m pytest core/tests -qpasses (nocore/changes in this PR)python skills/_registry/build_manifest.py skillsis clean (this PR doesn'ttouch the product's own
skills/registry —.claude/skills/is a separate,Claude-Code-only tree)
skills/<component>/<name>/scripts/check.pyadded or changedCLAUDE.mdrewritten,.bob/rules/01-cap-evolve.mdpoints atthe new skills,
scripts/offload-review/NOTICE.mdexplains the portbash scripts/offload-review/self_check.shpasses (new, not in CI yet):usage/validation errors, every backend defines the 3 required functions, every
shell script parses
lychee --offlineoverCLAUDE.md+ the new skill/script markdown: 0 errors🤖 Generated with Claude Code