메뉴
Autoresearch is the upstream meta-framework (Udit Goenka, MIT) for autonomous goal-directed iteration with safety guardrails. Locally split into 4 specialized skills. Start here to learn the pattern, then route to the right specialized skill.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/hoanghd218/ZaloCRM --skill ak-autoresearch명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Search library/framework documentation via llms.txt (context7.com). Use for API docs, GitHub repository analysis, technical documentation lookup, latest library features.
Interview-driven advisory skill. Analyzes a prompt or URL (GitHub issue, spec, doc), scouts the codebase, interviews the user one question at a time to reframe the problem into exact requirements and goals, then delivers honest advice: what to do, what to avoid, better alternatives, benefits and trade-offs. Use when the user asks for advice, a second opinion, a sanity check, requirement reframing before committing to a direction, or wants an existing plan or design pressure-tested (grilled) one question at a time.
Browser and desktop automation through the agent-browser CLI. Use for long autonomous browsing, compact page snapshots, screenshots, form filling, login/test flows without real Chrome profile state, scraping, exploratory QA, cloud browsers, Electron apps, Slack automation, and bug hunts.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | ak:autoresearch |
| description | Autoresearch is the upstream meta-framework (Udit Goenka, MIT) for autonomous goal-directed iteration with safety guardrails. Locally split into 4 specialized skills. Start here to learn the pattern, then route to the right specialized skill. |
| user-invocable | true |
| when_to_use | Invoke to route autonomous research loops to the right variant. |
| category | utilities |
| keywords | ["autoresearch","autonomous","iteration","karpathy","framework","lineage","router","umbrella"] |
| related | ["ck-loop","ck-predict","ck-scenario","ck-security"] |
| metadata | {"author":"agentkit","attribution":"Concept anchor for the autoresearch family by Udit Goenka (MIT), inspired by Karpathy's autoresearch pattern.","license":"MIT","version":"2.0.0"} |
Autoresearch is a framework, not a single skill. This page is a router.
The autoresearch pattern — autonomous goal-directed iteration with safety guardrails — is implemented across multiple AgentKit skills, each absorbing a different upstream sub-command. Use this page to find the right one, understand the lineage, and see what's not yet absorbed.
uditgoenka/autoresearch (MIT, ~4.2k stars)The upstream framework ships 11 sub-commands. Locally, 6 are absorbed (4 standalone skills + 2 folded as chain modes) and 5 are not yet absorbed. For the core loop, use /ak:loop.
| Upstream sub-command | Local skill | Status | Use when... |
|---|---|---|---|
/autoresearch (core) | /ak:loop | Faithful | Improving a measurable metric (coverage, bundle size, perf) over N bounded iterations |
/autoresearch:predict | /ak:predict | Faithful (in scope) | Multi-persona debate before risky changes; supports --chain reason and --chain probe (closed in #728) |
/autoresearch:scenario | /ak:scenario | Faithful (in scope) | Edge-case generation across 12 dimensions; supports both one-shot and iterative saturation (closed in #729) |
/autoresearch:security | /ak:security | Faithful (in scope) | STRIDE + OWASP audit with --fix; supports both one-shot and red-team-personas iterative discovery (closed in #730) |
/autoresearch:reason | /ak:predict --chain reason |
| Folded |
Subjective refinement loop — folded into /ak:predict chain modes rather than shipped as a standalone skill (closed in #728) |
/autoresearch:probe | /ak:predict --chain probe | Folded | Requirement interrogation — folded into /ak:predict chain modes rather than shipped as a standalone skill (closed in #728) |
/autoresearch:plan | — | Missing | Backfill candidate (HIGH priority) |
/autoresearch:debug | — | Missing | Backfill candidate (HIGH priority) |
/autoresearch:fix | — | Missing | Backfill candidate (MEDIUM); partly covered by /ak:fix |
/autoresearch:ship | — | Missing | Backfill candidate (MEDIUM); partly covered by /ak:ship |
/autoresearch:learn | — | Missing | Backfill candidate (LOW) — autonomous docs generator |
Drift assessments and backfill priorities come from the integration audit at plans/reports/researcher-260502-2145-autoresearch-integration-audit.md.
Almost never. This page is a discovery aid. To actually do work, route to one of the specialized skills above.
The exception: if you're authoring a new AgentKit skill that should adopt the autoresearch pattern (atomic commits, mandatory verify, guarded rollback), read this page + /ak:loop's references/ for the canonical implementation, then absorb only the relevant sub-command's workflow from upstream.
The autoresearch pattern grants the agent broad iterative authority — read, edit, run shell, commit. Every absorbed skill MUST honor:
experiment: prefix; each kept change committed, each discard git revert-clean.Verify — nothing kept unless verify exits >=0 and produces a measurable number. Failed verify = automatic rollback.Guard — when set, broken guard reverts the change. Use for "do not regress tests" / "do not break build."rm -rf /, fork bombs, fetch-and-execute (curl ... | sh), embedded credentials, unannounced outbound writes.Iterations: N over unbounded loops.Specialized skills document their concrete enforcement of these guardrails in their own references/.
/ak:loop/ak:predict/ak:scenario/ak:security.maintainer/external-sources.json (mode b_pattern_derived)Earlier versions of this skill shipped as a byte-identical copy of /ak:loop's body, marked "deprecated alias." That misrepresented the integration: locally the framework was already split, but the umbrella name absorbed nothing unique. As of 2026-05-02 (epic #711), this page is rewritten as a concept anchor so:
/ak:security's reference to "the ak:autoresearch pattern" lands somewhere meaningful.