Claude Cursor Skill

fixing-code

Imported from alexei-led/cc-thingz/dist/claude/dev-flow/skills/fixing-code.

LLM Mart · 0 points · 2 views 0 listing impressions 0 install-command copies
Download alexei-led-cc-thingz-dist_claude_dev-flow_skills_fixing-code-0202df8.zip · 1 KB

Install

skills CLI npx skills add https://github.com/alexei-led/cc-thingz/tree/master/dist/claude/dev-flow/skills/fixing-code
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install alexei-led-cc-thingz@llmmart
Git git clone https://github.com/alexei-led/cc-thingz.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole alexei-led/cc-thingz collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Fix and Diagnose Code

Fix the requested defect or failing gate, one verified root cause at a time. Use the matching writing-<lang> skill for toolchain commands. The optional coding-posture plugin owns general debugging and fix procedure; this skill owns the defect evidence and regression contract. Do not load both just to repeat process steps.

Without write access, return proposed changes (file, change, reason) instead of applying them.

Hard rules

  • Reproduce the failure when feasible. If it cannot be reproduced, patch only an evidenced contract violation; add regression coverage and report the reproduction gap. Otherwise ask for the missing artifact rather than patching a guess. Ask before adding temporary instrumentation.
  • No destructive git (hard reset, clean, force push, checkout over local changes) and no --no-verify.
  • Never disable assertions, skip fast tests, lower lint severity, or ignore files to make a check pass or run faster. Clear caches only when a stale cache is the diagnosed cause.
  • Fix the requested failure first; ask before expanding to unrelated failures.

Outcome

  • Repro: the fastest reliable failing signal, best a failing test at the behavior seam; otherwise a script, replayed payload, or small harness around the real path. If unavailable, report the gap and use evidenced contract violations instead.
  • Root cause: traced from the failing boundary to the first bad state or contract mismatch, with evidence (file:line, symptom, tool).
  • Regression test at the seam where the user saw the bug. If only a shallow seam exists, say so rather than adding a fake-confidence helper test.
  • Remove temporary probes or promote them to regression tests.
  • Browser-only symptoms go to browser-automation unless a cheaper CLI or unit signal exists.
  • If a code-graph tool (GitNexus, codegraph) is installed and fresh, use it to find callers or impact before changing widely used code.

Done when the original repro passes, or its absence is reported with the contract evidence and a passing regression test. Report a missing test seam; without a repro, stop rather than patch when regression coverage is unavailable. Relevant build/test/lint checks must pass, or each check that did not run must be named with its reason.

Report

Root cause with evidence, changes (path:line — fix), and each check with pass, fail, or skipped and why. If blocked, name the missing artifact, access, or approval.

Platform additions

Arguments

$ARGUMENTS may include:

  • diagnose or investigate: write the ranked hypotheses before any edit.
  • team: spawn read-only reviewer agents to challenge the root cause before editing. Ask each for root cause, evidence, suggested fix, and confidence; verify their claims yourself.

Claude tools

  • Ask for missing artifacts, access, or instrumentation approval with AskUserQuestion.
  • Track fixes longer than two steps with TaskCreate and TaskUpdate.
Files (cc-thingz)
  • SKILL.md 3.7 KB
    ---
    {"allowed-tools":["Task","TaskOutput","TaskCreate","TaskUpdate","TaskList","Read","Grep","Glob","Edit","Write","AskUserQuestion","Bash(make *)","Bash(go *)","Bash(golangci-lint *)","Bash(pytest *)","Bash(ruff *)","Bash(npm *)","Bash(bun *)","Bash(bunx *)","Bash(uv run pytest *)","Bash(cargo *)","Bash(dotnet *)","Bash(./gradlew *)","Bash(./mvnw *)"],"argument-hint":"[diagnose|investigate] [team]","context":"fork","description":"Fix code defects with a reproducible feedback loop, root-cause diagnosis, minimal patch, regression test, and clean verification. Use when debugging, diagnosing, or resolving lint/test/build failures. NOT for test-suite cleanup without a production bug (use improving-tests), or code review findings without fixes (use reviewing-code).","name":"fixing-code","user-invocable":true}
    ---
    
    # Fix and Diagnose Code
    
    Fix the requested defect or failing gate, one verified root cause at a time.
    Use the matching `writing-<lang>` skill for toolchain commands. The optional
    coding-posture plugin owns general debugging and fix procedure; this skill owns
    the defect evidence and regression contract. Do not load both just to repeat
    process steps.
    
    Without write access, return proposed changes (file, change, reason) instead of
    applying them.
    
    ## Hard rules
    
    - Reproduce the failure when feasible. If it cannot be reproduced, patch only an evidenced contract violation; add regression coverage and report the reproduction gap. Otherwise ask for the missing artifact rather than patching a guess. Ask before adding temporary instrumentation.
    - No destructive git (hard reset, clean, force push, checkout over local changes) and no `--no-verify`.
    - Never disable assertions, skip fast tests, lower lint severity, or ignore files to make a check pass or run faster. Clear caches only when a stale cache is the diagnosed cause.
    - Fix the requested failure first; ask before expanding to unrelated failures.
    
    ## Outcome
    
    - **Repro**: the fastest reliable failing signal, best a failing test at the behavior seam; otherwise a script, replayed payload, or small harness around the real path. If unavailable, report the gap and use evidenced contract violations instead.
    - **Root cause**: traced from the failing boundary to the first bad state or contract mismatch, with evidence (`file:line`, symptom, tool).
    - **Regression test** at the seam where the user saw the bug. If only a shallow seam exists, say so rather than adding a fake-confidence helper test.
    - Remove temporary probes or promote them to regression tests.
    - Browser-only symptoms go to browser-automation unless a cheaper CLI or unit signal exists.
    - If a code-graph tool (GitNexus, codegraph) is installed and fresh, use it to find callers or impact before changing widely used code.
    
    Done when the original repro passes, or its absence is reported with the
    contract evidence and a passing regression test. Report a missing test seam;
    without a repro, stop rather than patch when regression coverage is unavailable.
    Relevant build/test/lint checks must pass, or each check that did not run must
    be named with its reason.
    
    ## Report
    
    Root cause with evidence, changes (`path:line — fix`), and each check with
    pass, fail, or skipped and why. If blocked, name the missing artifact, access,
    or approval.
    
    ## Platform additions
    
    ### Arguments
    
    `$ARGUMENTS` may include:
    
    - `diagnose` or `investigate`: write the ranked hypotheses before any edit.
    - `team`: spawn read-only `reviewer` agents to challenge the root cause before editing. Ask each for root cause, evidence, suggested fix, and confidence; verify their claims yourself.
    
    ### Claude tools
    
    - Ask for missing artifacts, access, or instrumentation approval with `AskUserQuestion`.
    - Track fixes longer than two steps with `TaskCreate` and `TaskUpdate`.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.