ilandahan/AID / skills/autoresearch
autoresearch
VerifiedBounded improvement loop: one artifact, fixed eval score, keep an edit only if strictly better. Use for AR_DESIGN/AR_FUNCTION/AR_ACCEPTANCE stages or standalone bounded iteration.
| File | Kind | Size |
|---|---|---|
| SKILL.extended.md | markdown | 5.9 KB |
| SKILL.md | skill md | 21.6 KB |
| bin/check-harness-immutable.mjs | script | 1.2 KB |
| bin/check-harness-immutable.shexecutable | script | 1.9 KB |
| program.ar-acceptance.md | markdown | 15.3 KB |
| program.ar-design.md | markdown | 12.9 KB |
| program.ar-function.md | markdown | 17.0 KB |
| verify-run.shexecutable | script | 11.3 KB |
bin/check-harness-immutable.sh
View on GitHub#!/usr/bin/env bash
# check-harness-immutable.sh — skill-scoped PreToolUse(Write|Edit) for /autoresearch.
# REPO SOURCE OF TRUTH — installed to ~/.claude/skills/autoresearch/bin/ by
# tools/install-guard-hooks.js and drift-checked by tools/check-hooks-installed.js.
# The eval harness is FROZEN during an AR pass: editing criteria YAMLs, agent prompts,
# pipeline config or other skills makes the composite score meaningless and every
# keep/revert decision invalid. rubric_sha256 detects tamper post-hoc; this prevents it.
# Output contract: careful/freeze style — bare {"permissionDecision":...} or nothing.
# Fail-open on any parse failure. (No single quotes inside the python block.)
input=$(cat 2>/dev/null)
printf '%s' "$input" | python -c '
import json, re, sys
try:
d = json.load(sys.stdin)
path = (d.get("tool_input") or {}).get("file_path", "") or ""
except Exception:
raise SystemExit(0)
norm = path.replace("\\", "/")
# The runner MUST write these as part of its own loop (state.json is its mandated
# UNCONDITIONAL final action; results.tsv/snapshots live under autoresearch/) — never deny.
ALLOWED = [
r"\.aid/pipeline/state\.json$",
r"\.aid/pipeline/autoresearch/",
r"\.aid/pipeline/outer-counters\.sidecar$",
]
if any(re.search(p, norm) for p in ALLOWED):
raise SystemExit(0)
PROTECTED = [
r"/criteria/[^/]+\.ya?ml$",
r"/AGENT-PROMPT\.md$",
r"\.aid/[^ ]*\.json$",
r"\.mcp\.json$",
r"/\.claude/skills/",
]
if not any(re.search(p, norm) for p in PROTECTED):
raise SystemExit(0)
print(json.dumps({
"permissionDecision": "deny",
"message": "[autoresearch] Blocked: " + norm + " is part of the FROZEN eval harness "
"(criteria/prompts/pipeline config/skills). Editing it mid-pass makes the score "
"meaningless. Edit only the target artifact; harness changes happen outside AR runs.",
}))
' 2>/dev/null
exit 0
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |