neutralize
Skill0 starsChanged 58 days agoMITView on GitHub
Diagnose and fix nontrivial or recurring defects with reproduction, root-cause evidence, nearby-pattern scan, and a machine-readable gate. Trigger on repeated regression, build/runtime failure, logs or stack traces, "find the root cause", or "check for similar issues nearby". Obvious localized fixes use ordinary implementation.
In this repository22 other files it gives its agents.
- AGENTS.mdAGENTS.md
- acquisitionSkill
- after-actionSkill
- breachSkill
- cartographic-reconSkill
- chartSkill
- codifySkill
- conopsSkill
- coverSkill
- dossierSkill
- executeSkill
- go-nogoSkill
- herdrSkill
- netease-mailSkill
- paperworkSkill
- passdownSkill
- repairSkill
- sanitizeSkill
- taskingSkill
- tracebackSkill
- warroomSkill
- x-likes-digestSkill
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Use this workflow when at least one is true:
- Distinctive: line 31
Verify the exact symptom first, then the smallest relevant regression scope.
- Distinctive: line 32
Search adjacent files and similar call sites for the same error shape.
- Distinctive: line 33
If a public interface changed, record whether it expanded, stayed stable, or shrank and what complexity remains hidden behind it.
- Very specific: line 34
Write gate evidence file — failure_observation, reproduction_command or reproduction_skipped_reason, rerun_command, rerun_exit_code, …
- Distinctive: line 84
Do not patch from the symptom alone.
- Distinctive: line 85
Do not turn a bugfix into a broad refactor.
- Distinctive: line 86
Do not claim validation from a narrow command when the changed behavior requires a broader build/test.
- Very specific: line 87
If the same mistake appears in multiple places, either fix all confirmed instances or trigger codify/Canon incident so the pattern is not …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Fix Issue
Find the root cause, not just the symptom. Apply the smallest safe fix, then scan nearby code for the same mistake pattern.
Scope Gate
Use this workflow when at least one is true:
- root cause is unknown or spans multiple components
- the defect recurs or suggests a repeated mistake pattern
- build/runtime evidence, logs, stack traces, or screenshots must be reconciled
- the fix needs an auditable reproduction/rerun record
For a small, obvious local correction, fix and validate directly. Do not create
.agent-state/neutralize-gate.json merely because the user used the word "fix".
Workflow
- Extract evidence: exception type, file paths, line numbers, error text. For screenshots, read visible error text and identifiers first.
- Define the observable target: current failure evidence, expected result, and fastest rerun command. Reproduce if practical; otherwise record why reproduction is unavailable.
- Identify root cause.
- Apply the minimum safe correction as one verifiable slice.
- Verify the exact symptom first, then the smallest relevant regression scope.
- Search adjacent files and similar call sites for the same error shape.
- If a public interface changed, record whether it expanded, stayed stable, or shrank and what complexity remains hidden behind it.
- Write gate evidence file —
failure_observation,reproduction_commandorreproduction_skipped_reason,rerun_command,rerun_exit_code,adjacent_searches,adjacent_findings,public_interface_changed,boundary_assessment,remaining_risk→.agent-state/neutralize-gate.json. - Run gate:
blocked → 补证据后重跑。pass → 继续总结。python3 <skill-dir>/scripts/verify_fix_gate.py
Output
Human-readable summary:
- Observed Error / Root Cause / Fix Applied
- Similar Issues Checked / Additional Similar Issues Found
- Validation / Remaining Risk
Machine-readable evidence (.agent-state/neutralize-gate.json):
{
"failure_observation": "pytest fails with AssertionError at tests/test_foo.py:42",
"reproduction_command": "pytest tests/test_foo.py -v",
"reproduction_skipped_reason": "",
"rerun_command": "pytest tests/test_foo.py -v",
"rerun_exit_code": 0,
"rerun_output_sample": "...",
"adjacent_searches": ["rg 'same_pattern' src/", "rg 'similar_call' lib/"],
"adjacent_findings": ["src/bar.py:42 same pattern"],
"public_interface_changed": false,
"boundary_assessment": "public interface unchanged; correction remains inside parser module",
"remaining_risk": "only checked Python side; C++ callers not scanned"
}Rule Sources
- Canon
/media/yhr/2T/Canon/{projects,tasks,decisions,patterns,incidents}— primary source for durable context, known failure modes, reusable fixes. Query first. - Canon overrides
.agent-state/when both exist. Code overrides Canon when they conflict — Canon captures intent, code captures current reality. .agent-state/MEMORY.mdand.agent-state/rules/mistakes.md— compatibility fallback only.
Workflow Gate Contract
For bug fixes that produce code or durable lessons, follow the shared workflow output contract:
/home/yhr/.agents/repos/agent-skills/references/skill-output-contract.mdSmall local fixes may stop at validation. Complex fixes, repeated mistake patterns, build failures, and cross-project lessons must promote evidence to Canon before handoff.
Gotchas
- Do not patch from the symptom alone. If reproduction is impractical, record the strongest available evidence and the reason reproduction was skipped.
- Do not turn a bugfix into a broad refactor. Fix the minimum root cause first; adjacent cleanup is only allowed when it reduces the same failure pattern.
- Do not claim validation from a narrow command when the changed behavior requires a broader build/test. State the validation scope and remaining risk explicitly.
- If the same mistake appears in multiple places, either fix all confirmed instances or trigger
codify/Canon incident so the pattern is not lost.
Canon 输出边界
读取共享契约:/home/yhr/.agents/repos/agent-skills/references/canon-output-contract.md。
- Ordinary small fixes do not need Canon updates.
- Complex root causes, repeated mistake patterns, build failures, and cross-project lessons should create/update
/media/yhr/2T/Canon/raw/update-cards/<date>-neutralize-<topic>.md,incidents/, orpatterns/. - Validation logs, commits, screenshots, and local reports stay as artifact refs; do not copy them into Canon by default.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

