dankofly/perfectify / skill/dagx-agi-kernel
dagx-agi-kernel
VerifiedImprove and verify agent work after repeated failures, in dependency-heavy tasks, or when optimization claims need baseline and regression evidence. Use for DAGx/Perfectify requests, failed retries, risky multi-step work, or requests to verify an improvement. Exclude routine questions, drafting, one-step edits, and directly checkable calls.
playbook/playbook.md
View on GitHub# Perfectify Playbook (procedural memory, ACE format)
# Format: [id] helpful=N harmful=M :: rule (with trigger + test condition)
# Governance: retire when harmful >= helpful after >=5 trials; active cap 60; dedup semantically identical.
## gates
[gates-00001] helpful=3 harmful=0 :: Before ANY irreversible action: end turn with dry-run list plus one approval question. Trigger: delete/send/publish/purchase/shared-state-overwrite planned. Test: no mutation occurred before user reply.
[gates-00002] helpful=1 harmful=0 :: Before bulk deletion, produce a machine-checked dry-run with counts, cutoff boundary, and edge-case check (near-threshold records kept). Trigger: batch mutation of records. Test: dry-run output lists exact target count and at least one boundary case before any write.
## verification
[ver-00001] helpful=1 harmful=0 :: Round both outputs to the comparison precision before diffing float results. Trigger: numeric output equality check. Test: rounded-identical counts as pass.
[verificati-00001] helpful=1 harmful=0 :: Test the fix against the ORIGINAL failing input, not only new cases.
[verificati-00002] helpful=0 harmful=0 :: Verify selection criteria against both directions: targets matched AND near-miss records confirmed kept. Trigger: filter-based destructive operation. Test: kept-side sample inspected, not only deleted side.
[verificati-00003] helpful=0 harmful=0 :: Prove 'no slowdown' with a matched timing comparison against the un-fixed path (same run count, same machine), reporting overhead per run in ms, not just total wall time. Trigger: performance-nonregression claim for a fix that adds retries/backoff. Test: measured delta vs no-fix baseline is reported and negligible relative to suite time.
[verificati-00004] helpful=0 harmful=0 :: Fix the EARLIEST failing check first and rerun the full suite after each single fix, so later failures are observed on the already-partially-fixed state rather than batched. Trigger: multi-check suite with ordered iteration loop. Test: suite rerun logged after each individual fix; no two fixes applied before one verification.
## failure-recovery
[fail-00001] helpful=1 harmful=0 :: On intermittent failures, measure residual probability per run and size retries so P(fail) < 1/1000. Trigger: flaky test with known per-call failure rate p. Test: 60+ consecutive green runs recorded.
## governance
[govern-00001] helpful=0 harmful=0 :: Reject benchmark/task-specific bullets at ADD time; a rule naming specific ports, repos, or datasets violates generalization. Trigger: proposed bullet references concrete environment specifics. Test: bullet text contains no environment-unique identifiers.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |