maker-checker-relay
Skill0 starsChanged 4 months agoMITView on GitHub
Relay a Generator (implementer = Maker) and a read-only Verifier (Checker — the kit's dev-reviewer agent, a /review session, or a human) until tests are green AND the Checker has zero blocking findings. A goal-loop wrapper whose stop condition is a deterministic gate, not the model's self-report. Triggers: maker-checker, generator-verifier relay, implement-then-review loop, no self-review, /review loop, relay.
In this repository46 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- a11y-standardsSkill
- a11y-static-gateSkill
- adversarial-refute-voteSkill
- antigravity-buildSkill
- architecture-parity-gateSkill
- ci-required-check-installerSkill
- codex-buildSkill
- codex-imagegenSkill
- codex-spark-delegateSkill
- commit-convention-gateSkill
- commit-msg-quality-scoreSkill
- consumer-finding-ledgerSkill
- context-bloat-trackerSkill
- coverage-floor-lockSkill
- destructive-env-guardSkill
- dev-standardsSkill
- eval-dataset-buildSkill
- eval-regression-guardSkill
- fanout-verify-synthSkill
- gate-forgeSkill
- hallucination-rate-meterSkill
- harness-kpi-ledgerSkill
- judge-rubric-voteSkill
- knowledge-citation-coverageSkill
- knowledge-dedup-scanSkill
- knowledge-index-guardSkill
- live-state-verify-guardSkill
- memory-distillSkill
- mistake-rule-distillSkill
- oneshotSkill
- prompt-ab-evalSkill
- report-shapeSkill
- retry-failure-classifierSkill
- review-finding-reproSkill
- rule-dedup-scanSkill
- rule-promote-auditSkill
- rule-promotion-extractSkill
- secret-scan-hardeningSkill
- self-heal-ciSkill
- session-length-compact-auditSkill
- startSkill
- subagent-usage-analyzerSkill
- task-success-rate-trackSkill
- token-codegen-gateSkill
- trigger-vocab-dedupSkill
The file
Files
maker-checker-relay
A wrapper over scripts/goal-loop.sh (the stop primitive) that separates the Generator
(implementer = Maker) from the Verifier (a read-only reviewer = Checker) and relays
between them until BOTH hold:
- Maker's tests are green (a deterministic test command exits 0), and
- the Checker has zero blocking findings (a
review.outwith noBLOCKERlines).
This is the kit's Generator-Verifier principle applied to the fix loop. The point is to
remove the self-review bias — an implementer who reviews their own diff and declares it
"fixed". "Done" is decided by a deterministic gate's exit code, never by self-report; the
loop is bounded by --max (default 3) so it always terminates.
The Checker role
The Checker is any read-only reviewer that writes findings to a file — in this kit the
natural fit is the dev-reviewer agent (spawn it read-only on the diff), but a
/review session or a human works identically. What matters is that it runs in a separate
context from the Maker and marks each blocking finding with a line containing BLOCKER
(non-blocking nits are left unmarked).
How to run
- Decide the Maker DoD (a deterministic test command, exit 0 = green) and the Checker's
findings file
review.out. - Advance the relay one step at a time (call at the top of each cycle, branch on exit code):
bash "${CLAUDE_PLUGIN_ROOT:-.}/scripts/maker-checker-relay.sh" --tick \ --test "<Maker's deterministic test>" \ --review-out review.out [--blocker-pattern BLOCKER] \ [--goal "<desc>"] [--max 3] [--state .goal-loop-maker-checker.state]- exit 1 (🔁 CONTINUE):
- Maker (implementer): make ONE increment toward green tests. Do not self-review.
- Checker (read-only reviewer — the
dev-revieweragent,/review, or a human): review the diff, write findings toreview.out, and putBLOCKERon each blocking finding's line (omit nits). Then re-run--tick.
- exit 0 (✅ COMPLETE): tests green AND zero
BLOCKERlines. Commit / PR. - exit 2 (🛑 CAP): over
--max. Stop and escalate to a human; record the unmet goal wherever you track blockers.
- exit 1 (🔁 CONTINUE):
- Reset the attempt counter with
--resetwhen starting a new relay. - To wire it into goal-loop by hand,
--print-checkprints the--checkstring (already pointed at--gate, so the pass/fail logic is defined once).
Deterministic --check (self-test)
bash "${CLAUDE_PLUGIN_ROOT:-.}/scripts/maker-checker-relay.sh" --self-test- exit 0: the wrapper is healthy — verified
bash -n, that it referencesgoal-loop.sh, the stop wiring (--reset/--check/--max/--state), the pass/fail truth table (7 cases: green+clean = pass; green+blockers, red+clean, red+blockers, green+not-reviewed, no-test-command, and a malformed--blocker-pattern= fail — the gate fails CLOSED when its own matcher errors), and--print-checkend-to-end (2 cases), all on hermetic fixtures (mktemp -d, cleaned up — no repo residue). - non-zero: a gap in the pass/fail logic, wiring, or syntax.
When to use / not
- ✅ A deterministic Maker DoD exists (green tests) and independent review matters.
- ❌ No deterministic test, or a single-shot change that needs no iteration (use
/build).
Related
- primitive:
scripts/goal-loop.sh(exit 0=GOAL MET / 1=CONTINUE / 2=CAP REACHED) - wrapper:
scripts/maker-checker-relay.sh(--gate/--print-check/--tick/--reset/--self-test) - Verifier role: the
dev-revieweragent (read-only), or a/reviewsession - command:
/goal-loop(commands/goal-loop.md); profile:docs/harness-profile.md
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
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.

