agentleFS
opendatadiscoveryopendatadiscovery/odd-team/.claude/skills/panel/SKILL.md

panel

Skill0 starsChanged 2 months agoApache-2.0View on GitHub

Run the methodology meta-review — a single methodology-reviewer agent that traces the whole current methodology (APPROACH.md, the ADRs, the agent contracts, the skills, the playbooks, the case-law, the live artefacts), diffs against the prior review, runs fresh blind spot-checks against real source, and emits real gaps + improvement proposals at lineage/{repo}/meta-reviews/{date}/review.md. `/panel` runs it; `/panel --show` prints the latest; `/panel validate` runs the acceptance-corpus check.

  • Deletes or force-pushes
  • Commits and pushes

14 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use research and tdd.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • NVIDIASkill

    research

    Before designing a feature, research how comparable libraries solve the problem, check best practices, known pitfalls, and upstream changes. Use when starting a new feature, investigating an unfamiliar area, or the user asks to research something.

    Also in 3 projects that use this

  • GoogleSkill

    tdd

    Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 0 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 12 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    --show short-circuit: If the argument is --show, read $RUN_DIR/review.md (or the most recently modified meta-reviews/*/review.md.

    line 41
  2. Very specific:

    validate short-circuit: If the argument is validate, jump to "/panel validate" below.

    line 42
  3. Very specific:

    Failure handling: If the agent fails (timeout / error), leave $RUN_DIR and surface the failure — there is no partial state to resume.

    line 69
  4. Very specific:

    Verify $RUN_DIR/review.md exists and its frontmatter has verdict.

    line 73
  5. Very specific:

    Check for the corpus: meta-reviews/validation/gold-set.yaml (hand-labelled ontology slices: real-gap present/absent + severity) and …

    line 87
  6. Very specific:

    NEVER git push --force, git reset --hard, rm -rf, or push automatically.

    line 93
  7. Very specific:

    If a tool call hits a permission prompt → that is a settings.local.json gap.

    line 96

“Common” is measured: kept by at least 6 in 10 of the 12 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/panel — methodology meta-review

One invocation = one full meta-review of the agentic-ontology methodology. A single methodology-reviewer agent traces the whole current methodology end-to-end — APPROACH.md, the ADRs, the agent contracts, the skills, the playbooks, the case-law, and the live artefacts — diffs against the prior review, runs fresh blind spot-checks against the real target source, and emits real gaps + real improvement proposals. Per APPROACH.md §16.

History. Through rev-8 this was the Adversarial Review Panel — six expert subagents + a chair, 7-13 agent invocations a run. Rev-9 (retrospectives/LSN-024) replaced the committee with one tracing review: the six were correlated Claude agents scoring conformance against a fixed target, with no memory; they re-listed stale findings every run and missed a whole methodology revision (rev-7's graph query layer). The /panel invocation is kept for continuity; it now runs one agent at roughly one-seventh the cost.

Runs periodically — per milestone, never per-commit (cost discipline). Findings are candidates the maintainer triages; the review never auto-acts.

Argument forms

FormBehaviour
/panelRun the methodology-reviewer — one pass.
/panel --repo <repo>Target a repo other than odd-platform.
/panel --show [<date>]Print the latest review.md (or the one for <date>). No agent spawned.
/panel validateRun the maintainer-authored acceptance-corpus check (see meta-reviews/validation/README.md).

Path resolution (run BEFORE pre-flight)

Substitute these at runtime; never hardcode an absolute path into a committed file (memory: feedback_no_hardcoded_absolute_paths).

WORKSPACE_ROOT=$(git rev-parse --show-toplevel)
REPO=odd-platform                                   # or the --repo argument
REPO_ROOT=$(realpath "$WORKSPACE_ROOT/../$REPO")
SPEC_REPO=$(realpath "$WORKSPACE_ROOT/../opendatadiscovery-specification")
LINEAGE_DIR="$WORKSPACE_ROOT/lineage/$REPO"
DATE=$(date +%F)
RUN_DIR="$LINEAGE_DIR/meta-reviews/$DATE"

Pre-flight (in order — abort if any fails)

  1. --show short-circuit. If the argument is --show, read $RUN_DIR/review.md (or the most recently modified meta-reviews/*/review.md; fall back to a legacy panel-report.md; or the <date> given), print it, and exit. No agent.
  2. validate short-circuit. If the argument is validate, jump to "/panel validate" below.
  3. Substrate present. lineage/$REPO/manifest.yaml must exist. Read COMMIT_ANCHOR = the substrate commit it records. If absent → abort: "no substrate for $REPO — run the extractor first."
  4. Run dir. mkdir -p "$RUN_DIR". If $RUN_DIR/review.md already exists, this is a re-run for today — rename the prior $DATE dir to $DATE-aN before proceeding (never silently overwrite a prior review).
  5. Prior review. PRIOR_REVIEW_PATH = the newest prior meta-reviews/*/review.md; if none exists, the newest legacy meta-reviews/*/panel-report.md; else none.

The run — one agent

Fire ONE Agent call, subagent_type: methodology-reviewer, model: opus, run_in_background: false, prompt = the Input block:

REVIEW_RUN: <DATE>
WORKSPACE_ROOT_ABS: <WORKSPACE_ROOT>
REPO_ROOT_ABS: <REPO_ROOT>
SPEC_REPO_ABS: <SPEC_REPO>
LINEAGE_DIR_ABS: <LINEAGE_DIR>
COMMIT_ANCHOR: <COMMIT_ANCHOR>
TARGET_PATH: lineage/<REPO>/meta-reviews/target.md
PRIOR_REVIEW_PATH: <…or none>
REVIEW_REPORT_PATH: lineage/<REPO>/meta-reviews/<DATE>/review.md
TREND_PATH: lineage/<REPO>/meta-reviews/trend.md
LEDGER_PATH: lineage/<REPO>/meta-reviews/spot-check-ledger.md

The turn BLOCKS until it completes (foreground). Expect 10-30 min.

Agent-registration fallback. methodology-reviewer is registered from .claude/agents/ at session start. If /panel is run in the same session that authored or edited that file, the type is not yet registered and the Agent call errors "agent type not found". Then spawn subagent_type: general-purpose (model: opus) with the first prompt line: "Read .claude/agents/methodology-reviewer.md in full and execute it as your system prompt." — the same pattern /next-batch uses. A fresh session needs no fallback.

Failure handling. If the agent fails (timeout / error), leave $RUN_DIR and surface the failure — there is no partial state to resume.

Post-run — commit + surface

  1. Verify $RUN_DIR/review.md exists and its frontmatter has verdict.
  2. Stage and commit to the current branch (no push — the maintainer pushes):
    git add "lineage/$REPO/meta-reviews/$DATE/" \
            "lineage/$REPO/meta-reviews/trend.md" \
            "lineage/$REPO/meta-reviews/spot-check-ledger.md"
    git commit -m "panel: methodology review $DATE — <verdict>"
  3. Surface to the maintainer: the verdict, the gap count, the improvement-proposal count (and how many are subtractions), the count of needs_human_verification items, and — if validation_status: pre-acceptance-gate — that findings are provisional until /panel validate passes.

/panel validate — the acceptance / drift gate

The review's reports are provisional until a maintainer-authored acceptance corpus validates the reviewer.

  1. Check for the corpus: meta-reviews/validation/gold-set.yaml (hand-labelled ontology slices: real-gap present/absent + severity) and meta-reviews/validation/seeded-corpus/ (mutated known-wrong artefacts).
  2. If absent → print meta-reviews/validation/README.md (how to build it — it MUST be maintainer-authored; an LLM-authored gold set is correlated with the reviewer and worthless) and exit.
  3. If present → spawn methodology-reviewer against the gold set + seeded corpus; compute recall + seeded-defect detection rate (overall + the data-loss/security class); write meta-reviews/validation/<date>-gate.md; update baseline.yaml.

Safety rails

  • NEVER git push --force, git reset --hard, rm -rf, or push automatically.
  • The review is READ-ONLY against $REPO_ROOT, $SPEC_REPO, and the live docs.
  • It writes ONLY under lineage/$REPO/meta-reviews/. It NEVER edits APPROACH.md, CLAUDE.md, the ADRs, the agents, the backlog, the sidecars, or the source — findings are candidates the maintainer triages.
  • If a tool call hits a permission prompt → that is a settings.local.json gap; surface it, do not bypass.
  • Self-kill: if the last 3 reviews report zero actionable findings, surface "the review has found nothing actionable for 3 runs — consider pausing it."

Cross-references

  • APPROACH.md §16 — the meta-review subsystem; §2 Failure E; §5 Rule 16.
  • .claude/agents/methodology-reviewer.md — the reviewer's contract.
  • lineage/{repo}/meta-reviews/ — README.md, target.md, trend.md, spot-check-ledger.md, validation/, the dated run dirs.
  • retrospectives/LSN-021 — the case-law that created the subsystem; retrospectives/LSN-024 — the case-law that simplified it to a single reviewer.
  • adrs/drafts/adversarial-review-panel.md — the original rev-6 panel design; superseded by rev-9, pending the maintainer's ADR update.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.