메뉴
Runtime helper for autonomous-research-loop ledgers. Use to initialize, append, validate, inspect, or smoke-test autonomous research loop state files without network, package installation, provider CLI calls, or live agent spawning.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when the user asks for a multi-agent discussion, panel review, multi-agent review, or multi-agent research session with role selection, round control, and template-based orchestration.
Use when drafting, validating, or normalizing bounded cross-agent task/result packets for parent-controlled handoffs. This is a packet-contract skill, not a runtime delegation broker.
Send email over SMTP using only the Python standard library, with plain-text and HTML bodies, file attachments, cc/bcc, reply-to, a dry-run preview, connection verification, and redacted config inspection.
Run bounded autonomous research iterations with evidence gates, recovery ledgers, and optional cross-agent handoffs. Use when the user asks to continue research autonomously, run a research loop, integrate autonomous agent loops, or keep improving a research workflow without repeated prompts.
Use for external DOI/ISBN/title resolution, manifest creation from pasted text, and paper retrieval after the local library-first workflow does not satisfy the request or the user explicitly opts out of library use.
Use when a research task benefits from an explicit phased workflow with structured source handoff across search, analysis, and writing, and when preserving citations across phases matters.
| name | autonomous-research-loop-runtime |
| description | Runtime helper for autonomous-research-loop ledgers. Use to initialize, append, validate, inspect, or smoke-test autonomous research loop state files without network, package installation, provider CLI calls, or live agent spawning. |
This companion skill provides offline helper scripts for the
autonomous-research-loop ledger contract.
It is intentionally runtime-backed and should be installed only for targets that support runtime skill helpers. It is not an OpenClaw skill-file target.
From a configured ai-agents-skills runtime, prefer:
bash "$AAS_RUNTIME_ROOT/run_skill.sh" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.sh selftest
Common commands:
bash "$AAS_RUNTIME_ROOT/run_skill.sh" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.sh init --dir research/run --goal "..." --success-criteria "..."
bash "$AAS_RUNTIME_ROOT/run_skill.sh" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.sh append-iteration --dir research/run --mode bounded-research --objective "Check evidence gaps" --decision continue
bash "$AAS_RUNTIME_ROOT/run_skill.sh" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.sh validate --dir research/run
bash "$AAS_RUNTIME_ROOT/run_skill.sh" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.sh status --dir research/run
The helper is authoritative for local ledger and iteration-budget invariants.
It rejects appends after max_iterations, rejects continuing decisions on the
final allowed iteration, rejects early stop records that lack a valid
proof/success artifact, and validation fails ledgers whose spent iteration
count, iteration records, terminal decisions, and running status disagree.
The runtime also exposes force-management and enforcement subcommands used by the
autoloop wiring (not part of the normal ledger flow): arm / disarm /
active register, deregister, and list an active loop; done is the read-only
stop-condition arbiter; hook-check is the cross-platform Stop-hook check that
the installed Claude hooks.Stop entry invokes directly (it reads the hook JSON
on stdin, honors AUTOLOOP_DISABLE / AUTOLOOP_DRIVER / the stop_hook_active
re-entrancy payload, and exits 2 only when an active loop is unfinished, fail-open
otherwise); agent-cmd prints the per-provider headless one-iteration command
(offline PATH probe, no execution); and drive is the cross-platform headless
driver that runs one iteration per loop until done (the POSIX
autoloop_driver.sh is a thin shim that delegates to it).
A chat session cannot run hundreds of loop iterations: context windows and turn
boundaries end it. Unattended execution therefore uses drive, which respawns a
FRESH headless agent session per iteration against the on-disk loop files and
owns the stop conditions itself. Exactly one of --cmd or --provider selects
the iteration command; with --provider the runtime builds the standard
one-iteration invocation for that install target:
| Provider (target) | Iteration command built by agent-cmd / drive --provider |
|---|---|
claude | claude -p "<prompt>" --dangerously-skip-permissions |
codex | codex exec --full-auto "<prompt>" |
deepseek | codewhale exec --auto "<prompt>" (falls back to codewhale-tui, deepseek) |
opencode | opencode run "<prompt>" |
copilot | copilot -p "<prompt>" --allow-all-tools |
antigravity | gemini --yolo -p "<prompt>" |
<prompt> is the standard one-iteration contract: read recovery.md and the
ledger, execute the single recorded next action under the loop policy, verify
independently, append exactly one iteration record, refresh the recovery files,
exit. Inspect it with agent-cmd --provider <p> --dir <loop> --print-prompt.
OpenClaw is not a driver target (no local agent CLI); drive its loops from a
supported provider instead.
Start an unattended run (POSIX):
bash "$AAS_RUNTIME_ROOT/run_skill.sh" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.sh drive --dir research/run --provider claude
On Windows use %AAS_RUNTIME_ROOT%\run_skill.bat ... run_autonomous_research_loop.bat drive --dir research\run --provider codex.
Wrap with nohup, systemd-run, or Task Scheduler for multi-day runs.
Driver behavior:
<loop>/driver_logs/.--quota-backoff seconds (default 900) and retries, honoring the
pause-and-wait-for-credits policy. --max-quota-waits N caps consecutive
waits (default 0 = wait indefinitely).--max-failures consecutive occurrences.done arbiter: iteration
cap, wall/token/USD budgets, terminal ledger status, STOP_REQUESTED and
PAUSE sentinels, and require_user_stop_only.done), 3 max failures, 4 runtime error,
5 quota waits exhausted, 6 provider binary unavailable.AAS_AUTOLOOP_BIN_<PROVIDER> (binary), AAS_AUTOLOOP_ARGS_<PROVIDER>
(argument template; {prompt}/{dir} placeholders), AAS_AUTOLOOP_CMD_<PROVIDER>
(full shell template; {prompt} is inserted shell-quoted and also exported as
AUTOLOOP_PROMPT).The default flag sets grant the agent full tool autonomy, which unattended
research requires; run loops only in workspaces you trust the agent to modify,
and prefer a dedicated project root. Interactive forcing is separate: on Claude
the installed hooks.Stop entry blocks turn-end while an ARMED loop (arm --dir <loop> --root <project>) is unfinished; the other targets have no Stop
hook and are governed by the driver alone.
For an early proof/success stop, at least one --evidence-id ID must resolve to
proof_artifacts/ID.json inside the loop directory. Early proof/success stop
reasons are success, success_criteria_met, proof, proof_found,
found_proof, and proved. The artifact id must be 1-128 characters of
letters, digits, underscore, hyphen, or dot and must start with a letter or
digit. The JSON artifact must include:
{
"schema_version": "1.0",
"id": "proof-artifact-1",
"artifact_type": "lean",
"machine_checkable": true,
"target": "the theorem or success target",
"proof_path": "proofs/theorem.lean",
"checker": {
"name": "lean",
"status": "passed"
}
}
The helper checks that the artifact exists, id matches the evidence id,
schema_version is 1.0, machine_checkable is true, artifact_type is
one of lean, coq, isabelle, agda, sagemath, python-verifier, or
external-verifier, checker.name is non-empty, checker.status is passed,
target is non-empty, and proof_path is an existing relative file within
the loop directory. It does not run Lean, Coq, SageMath, or another checker
itself.
On Windows, use the installed runtime runner with the native launcher target:
%AAS_RUNTIME_ROOT%\run_skill.bat skills/autonomous-research-loop-runtime/run_autonomous_research_loop.bat selftest
& "$env:AAS_RUNTIME_ROOT\run_skill.ps1" skills/autonomous-research-loop-runtime/run_autonomous_research_loop.ps1 selftest
The helper:
driver_logs/)done, hook-check, agent-cmd, and selftest never
call Codex, Claude, Copilot, DeepSeek, or other provider CLIs; only drive
executes the iteration command the operator selected (via --cmd or
--provider), which is the entire point of the headless driverdrive may)Use the canonical autonomous-research-loop skill for orchestration policy and
this helper only for local ledger mechanics. This helper validates that an
early proof stop points to a passed machine-checkable proof artifact record; it
does not independently validate the semantic truth of the proof.
When this skill is involved, consider these workflow templates (install via
the workflow-templates artifact profile, or --with-deps to pull backing skills):
autonomous-research-loop-runbook -- Bounded autonomous research-loop runbook with four stop conditions, single-path solving, mandatory cross-agent verification, fresh-agent backtracking, and Modal/GitHub Actions credit-gated heavy-compute offload.