跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

Autoresearch

Claude Autoresearch Skill — Autonomous goal-directed iteration for Claude Code. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.

uditgoenka · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add uditgoenka/autoresearch

技能说明


name: autoresearch description: "Autonomous iteration loop: modify, verify, keep/discard against any metric" version: 2.2.2

Autoresearch — Autonomous Goal-directed Iteration

Safety Invariants (all subcommands)

  • Never push, publish, or deploy without explicit user approval.
  • Bounded by default. Override with Iterations: unlimited.
  • All results logged to autoresearch/{subcommand}-{YYMMDD}-{HHMM}/ directory.
  • Chain handoff via handoff.json. Evals reads *-results.tsv.

Dispatch (bare $autoresearch)

Parse the invocation in this order:

ConditionMode
Metric: or Verify: presentClassic — existing metric loop, unchanged
Free-form natural-language goal, no metric/verifyOrchestrator — see Orchestrator section
NothingSetup wizard — interactive config builder
--classic flagForce Classic regardless of goal text
--auto flagForce Orchestrator regardless of goal text

Print a banner on every invocation: [autoresearch] mode: classic | orchestrator | wizard.

Subcommands

CommandDoesDefault Iterations
$autoresearchIterate against a metric: modify → verify → keep/discard25
$autoresearch planConvert a goal into validated Scope, Metric, Verify configN/A
$autoresearch debugHunt bugs: hypothesize → test → falsify → repeat15
$autoresearch fixCrush errors one-by-one until zero remain20
$autoresearch securitySTRIDE + OWASP audit with red-team personas15
$autoresearch shipShip through 8 phases: checklist → dry-run → deploy → verifyN/A
$autoresearch scenarioGenerate edge cases across 12 dimensions20
$autoresearch predict5 expert personas debate before implementationN/A
$autoresearch learnScout codebase → generate docs or wiki → validate → fix loop10
$autoresearch reasonAdversarial debate with blind judges until convergence8
$autoresearch probe8 personas interrogate requirements until saturation15
$autoresearch improveResearch ICP challenges, discover improvements, generate PRDs15
$autoresearch evalsAnalyze iteration results: trends, plateaus, regressionsN/A
$autoresearch regressionRegression stability gate: baseline vs candidate, verdict STABLE/UNSTABLEN/A

Universal Flags

FlagApplies ToPurpose
Iterations: NAll loopingSet iteration count
Iterations: unlimitedAll loopingOpt-in unbounded
--evalsAll loopingMid-loop checkpoints + final summary
--evals-interval NAll loopingOverride checkpoint frequency
--chain <targets>AllSequential handoff after completion
--<subcommand>AllShorthand for --chain <subcommand>
--dry-runOrchestratorPrint derived config + planned pipeline; no execution
--max-cycles NOrchestratorHard ceiling on orchestration cycles (default 50)
--classicBare $autoresearchForce Classic metric-loop mode
--autoBare $autoresearchForce Orchestrator mode

Orchestrator

Activated when a plain-language goal is given without Metric:/Verify:. Classifies the goal into a Goal archetype — see references/orchestrator-routing.md for the archetype table and router decision table.

Resolve every scripts/... path below relative to this installed skill directory, never relative to the caller's working directory.

Two modes based on archetype:

  • Orchestration loop — predicate-bearing archetypes (ship-ready, optimize-metric, fix-broken, harden, build-feature, explore). Goal has a mechanical Success predicate; the loop runs until that predicate is met.
  • Single-pass dispatch — subjective/terminal archetypes (document, what-to-build, decide-design). Routes once to the fitting subcommand (learn / improve / reason), lets it self-terminate, then reports. No loop, no Plateau, no ship gate.

Orchestration Loop Steps

Backed by scripts/orchestrate.sh (deterministic seam — all routing logic lives there). Subcommands exposed: classify, next-hop, units, plateau, screen-cmd, verdict, validate-state, screen-state-predicate.

  1. Classify — scripts/orchestrate.sh classify "<goal>" → archetype label + mode.
  2. Derive predicate — reuse plan logic to produce a concrete Success predicate: exact shell command + expected output. For optimize-metric, run the full plan/wizard derivation internally.
  3. Confirm — ONE request_user_input showing: archetype, mode, concrete predicate (command + expected output), terminal choice (stop-at-verified vs proceed-to-ship). Misclassifications are caught here, not mid-run.
  4. Round-0 dry-run — prove the predicate command runs and returns a value; safety-screen every derived command via screen-cmd; print projected cycle budget. Stop here if --dry-run.
  5. Loop until predicate satisfied: a. Assess state via cheap signals (last handoff.json, regression verdict, error count) + affected-test verify. b. scripts/orchestrate.sh next-hop orchestrator-state.json → next subcommand. c. Run subcommand (its own bounded inner loop). d. Record per-hop outcome ∈ {progressed, no-op, failed, blocked}. e. Fold hop's handoff.json into orchestrator-state.json. f. scripts/orchestrate.sh units → recompute Units remaining.
  6. Stop conditions (checked after each hop):
    • Predicate met → ship gate (only if ship is in the pipeline) else CONVERGED.
    • scripts/orchestrate.sh plateau orchestrator-state.json → true → stop + report PLATEAU.
    • Cycles > ceiling (default 50, override --max-cycles N) → stop + report CEILING.
    • Hop outcome blocked/failed with no alternative route → checkpoint + stop + report BLOCKED.

Orchestrator State

orchestrator-state.json — orchestrator-owned, additive. Tracks: goal, archetype, predicate, terminal-choice, units_remaining

阅读完整来源 ↗