본문으로 건너뛰기
cccskills
무료GitHub에 공개

sharp-edges

Living catalogue of 7 known hazard entries (SE-01 through SE-07) specific to agent-studio: Windows backslash paths, prototype pollution, hook exit codes, async swallowing, ReDoS in glob-to-regex, DST arithmetic, array mutation during iteration. Each entry: symptom, root cause, fix, test assertion.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

설치 방법 보기

포함된 파일(10)

  • SKILL.md7.8 KB
  • commands/sharp-edges.md111 B
  • hooks/post-execute.cjs300 B
  • hooks/pre-execute.cjs417 B
  • references/research-requirements.md573 B
  • rules/sharp-edges.md609 B
  • schemas/input.schema.json766 B
  • schemas/output.schema.json255 B
  • scripts/main.cjs601 B
  • templates/implementation-template.md185 B

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Sharp Edges

Living catalogue of confirmed hazard patterns in agent-studio. Each entry documents a real bug we've shipped. Invoke this skill during debugging and code review to check against known failure modes.

SE-01: Windows Backslash Paths

Symptom: Glob patterns match correctly in CI (Linux) but silently fail on developer machines (Windows).

Root cause: path.relative() returns backslash-separated paths on Windows (node_modules\foo), but glob patterns use forward slashes. [^/]* in regex won't block \.

Fix:

// ALWAYS normalize before regex matching
const rel = path.relative(root, filePath).replace(/\\/g, '/');

Test assertion:

assert(normalizePath('a\\b\\c') === 'a/b/c');
assert(!normalizePath('a\\b').includes('\\'));

Files affected: .claude/lib/utils/path-constants.cjs, any glob-based exclusion logic.


SE-02: Prototype Pollution via JSON.parse

Symptom: Object.prototype gains unexpected properties after parsing malformed JSON from hook input or agent memory.

Root cause: JSON.parse('{"__proto__":{"isAdmin":true}}') silently mutates Object.prototype in older Node.js versions or when objects are created via Object.create(null) checks are missed.

Fix:

// Use safeParseJSON — strips __proto__, constructor, prototype keys
const { safeParseJSON } = require('.claude/lib/utils/safe-json.cjs');
const { success, data } = safeParseJSON(rawInput, {});

Test assertion:

const before = Object.getPrototypeOf({});
safeParseJSON('{"__proto__":{"evil":true}}', {});
assert(!{}.evil);

Files affected: Any hook that calls JSON.parse() on stdin input.


SE-03: Hook Exit Code Protocol

Symptom: Hook appears to "block" when it should allow, or silently "allows" when it should block.

Root cause: Hooks use exit codes 0 (allow) and 2 (block). Using process.exit(1) or process.exit(true) does not block — it produces unexpected behavior.

Fix:

// CORRECT
if (shouldBlock) process.exit(2); // Block tool execution
process.exit(0); // Allow tool execution

// WRONG
if (shouldBlock) process.exit(1); // Not a valid block code
process.exit(false); // Not valid

Test assertion:

// Mock hook execution and verify exit code
const result = execHook(input);
assert(result.exitCode === 0 || result.exitCode === 2);

Files affected: All hooks in .claude/hooks/.


SE-04: Async Exception Swallowing in Promise.all

Symptom: One failing async operation silently causes all results to be lost; agent believes all tasks succeeded.

Root cause: Promise.all([...]) rejects on the first failure and discards all other results. When wrapped in try/catch that returns [], the caller sees empty results with no error.

Fix:

// Use Promise.allSettled to get partial results
const results = await Promise.allSettled(tasks);
const succeeded = results.filter(r => r.status === 'fulfilled').map(r => r.value);
const failed = results.filter(r => r.status === 'rejected').map(r => r.reason);
return { succeeded, failed, partial: failed.length > 0 };

Test assertion:

const tasks = [resolvesWith('ok'), rejectsWith('error'), resolvesWith('ok2')];
const result = await safeAllSettled(tasks);
assert(result.succeeded.length === 2);
assert(result.failed.length === 1);

SE-05: ReDoS in Glob-to-Regex

Symptom: Glob pattern matching hangs or takes exponential time on long path strings.

Root cause: Naive glob-to-regex conversion produces patterns with nested quantifiers like (.*)* or (.+)+ which backtrack exponentially on non-matching strings.

Fix:

// Use anchored, non-backtracking patterns
// BAD: (.*)* -> exponential backtracking
// GOOD: [^/]* -> linear, no backtracking
function globToRegex(glob) {
  const escaped = glob.replace(/[.+^${}()|[\]\\]/g, '\\$&');
  return escaped
    .replace(/\*\*/g, '___DOUBLE___')
    .replace(/\*/g, '[^/]*') // Single * -> no path separators
    .replace(/___DOUBLE___/g, '.*'); // ** -> any path
}

Test assertion:

const regex = globToRegex('**/*.js');
const longString = 'a'.repeat(10000) + '.ts';
const start = Date.now();
regex.test(longString);
assert(Date.now() - start < 100); // Must complete in <100ms

SE-06: DST Arithmetic Bugs

Symptom: Date calculations are off by 1 hour for events that cross daylight saving time boundaries.

Root cause: Adding 24 60 60 * 1000ms to a timestamp does not always equal "tomorrow" — DST transitions can make a day 23 or 25 hours long.

Fix:

// WRONG: assumes 24h = 1 day
const tomorrow = new Date(date.getTime() + 86400000);

// CORRECT: use date arithmetic, not ms arithmetic
const tomorrow = new Date(date);
tomorrow.setDate(tomorrow.getDate() + 1);

Test assertion:

// Test across a known DST boundary (e.g., March DST change)
const dst = new Date('2026-03-08T01:00:00'); // Day before US spring forward
const nextDay = addDays(dst, 1);
assert(nextDay.getDate() === 9);
assert(nextDay.getHours() === 1); // Same time, next day

SE-07: Array Mutation During forEach Iteration

Symptom: Items are skipped or processed twice; behavior is non-deterministic.

Root cause: Mutating an array (push, splice, shift) while iterating with forEach or for...of causes the iterator to skip or re-visit elements.

Fix:

// WRONG: mutates during iteration
arr.forEach(item => {
  if (condition) arr.push(newItem); // Skips items
});

// CORRECT: collect mutations, apply after
const toAdd = [];
arr.forEach(item => {
  if (condition) toAdd.push(newItem);
});
arr.push(...toAdd);

// OR: work on a copy
[...arr].forEach(item => { ... });

Test assertion:

const arr = [1, 2, 3];
const result = safeForEach(arr, item => item * 2);
assert(result.length === 3); // Original not mutated, all items processed

Usage

Invoke at the START of any debugging session:

Skill({ skill: 'sharp-edges' });

Then match your symptom against SE-01 through SE-07. If your bug is new and reproducible, add it to this catalogue.

Adding New Entries

Pattern for new entries:

## SE-0N: [Short Title]
**Symptom:** [What you observe]
**Root cause:** [Why it happens]
**Fix:** [Code fix with before/after]
**Test assertion:** [Minimal test that would catch it]
**Files affected:** [Where in agent-studio this applies]

Memory Protocol

After invoking: if you find a new sharp edge, append it here via technical-writer agent.

Memory Protocol (MANDATORY)

Before starting: Read .claude/context/memory/learnings.md

After completing:

  • New pattern -> .claude/context/memory/learnings.md
  • Issue found -> .claude/context/memory/issues.md
  • Decision made -> .claude/context/memory/decisions.md

ASSUME INTERRUPTION: If it's not in memory, it didn't happen.

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Ensure accessibility in UI components including semantic HTML, ARIA attributes, keyboard navigation, and WCAG 2.2 AA compliance.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

oimiragieo/agent-studio422026년 7월 14일 업데이트

Use when you want to improve response quality through meta-cognitive reasoning. Applies 15+ reasoning methods to reconsider and refine initial outputs.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

oimiragieo/agent-studio422026년 7월 14일 업데이트

N-round opposing-stance debates for trade-off analysis. Assigns pro/con roles to agents, runs structured debate rounds with quality scoring, and produces a moderator synthesis with confidence-rated recommendation. Generalizable to architecture, technology, security, and design decisions.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

oimiragieo/agent-studio422026년 7월 14일 업데이트

Force adversarial code review stance that eliminates confirmation bias — reviewer must find issues or re-analyze

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

oimiragieo/agent-studio422026년 7월 14일 업데이트

Creates specialized AI agents on-demand when no existing agent matches a request. Use when the Router cannot find a suitable agent for a task. Enables self-evolution by generating persistent agents.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

oimiragieo/agent-studio422026년 7월 14일 업데이트

LLM-as-judge evaluation framework with 5-dimension rubric (accuracy, groundedness, coherence, completeness, helpfulness) for scoring AI-generated content quality with weighted composite scores and evidence citations

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

oimiragieo/agent-studio422026년 7월 14일 업데이트

oimiragieo의 스킬 모두 보기

이 스킬의 문제 신고