사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/security-talent/bugskill-ai --skill amend-skill시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Amend Skill
Purpose
Skills degrade over time as targets change, new bypass techniques emerge, and failure patterns accumulate. This skill closes the feedback loop: it reads execution history, identifies what is systematically failing, and proposes a minimal surgical amendment to the skill.
Trigger Conditions
Activate this skill when:
- User says "improve this skill", "fix this skill", "update this skill", "amend this skill"
- User says "why does X keep failing" where X is a skill name
- User says "add this to the skill" after describing a new technique that worked
observations/<skill-name>/runs.mdcontains 3 or more entries withoutcome: fail
Methodology
- Identify the target skill — confirm the skill name with the user if ambiguous.
- Read the skill — load
skills/<bucket>/<category>/<skill-name>/SKILL.mdin full. - Read the observations — load
observations/<skill-name>/runs.md. - Analyze failure patterns:
- Which steps consistently fail?
- Which payloads are blocked or produce wrong output?
- Which trigger conditions are missing or too broad?
- Are there new bypass techniques that worked but aren't documented?
- Is the output format causing downstream confusion?
- Formulate a single targeted amendment — one focused change rather than a full rewrite. The amendment should be one of:
- Tighten trigger: remove or refine a trigger condition that causes false activation
- Add missing condition: add a step, payload, or bypass that is now needed
- Reorder steps: move a high-value step earlier to avoid wasted time
- Fix payload: update an outdated payload that no longer works
- Change output format: restructure how the skill presents findings
- Output the amendment as a unified diff — show old text and proposed new text side by side.
- Ask for confirmation — do not apply until the user approves.
- Apply and record — edit the SKILL.md, then append to
observations/<skill-name>/runs.md:
---
date: YYYY-MM-DD
skill: <skill-name>
task: Amendment applied via amend-skill
outcome: amendment
what_worked: |
<describe what the amendment changes and why>
what_failed: |
<describe the failure pattern that motivated the amendment>
errors: |
N/A
notes: |
Amendment type: <tighten-trigger | add-condition | reorder-steps | fix-payload | change-output>
Rationale: <one-sentence reason>
---Example Interaction
User: "why does sql-injection keep failing on this target?"
Process:
- Read
skills/owasp-wstg/injection/sql-injection/SKILL.md - Read
observations/sql-injection/runs.md— finds 3 fail entries, all noting WAF blocking'and-- - Analysis: Bypass section exists but WAF-specific encoding variants not prominent enough in step 2
- Amendment: Move WAF bypass step earlier in Methodology, add a WAF detection pre-step
Output:
-2. Submit `'`, `"`, `;`, `--`, `/* */` individually and observe response differences.
-3. Confirm with boolean pair: append `AND 1=1--` (true) vs `AND 1=2--` (false).
+2. Check for WAF: submit `'` and observe whether the response is a WAF block page (uniform
+ error regardless of input) vs. a database error (input-specific). If WAF present, use
+ encoding variants from the Bypass Techniques section before confirming injection.
+3. Submit `'`, `"`, `;`, `--`, `/* */` individually. If WAF present, substitute with
+ `%27`, `%2D%2D`, and comment fragmentation `UN/**/ION`.
+4. Confirm with boolean pair: `AND 1=1--` vs `AND 1=2--`.Fix Patterns
This is a meta-skill — it has no fix patterns of its own. Its output is an amendment to another skill's fix patterns.