Claude Skill

diagnose

Debug coding failures through reproduction, ranked hypotheses, narrow fixes, and verification. Use when a bug, regression, flaky test, performance failure, or inconsistent behavior is broken, slow, throwing, or hard to reproduce and the task needs disciplined diagnosis instead of

LLM Mart · 0 points · 3 views 0 listing impressions 0 install-command copies
Download markoblogo-abvx-agent-skills-skills_diagnose-37a7146.zip · 2 KB

Install

skills CLI npx skills add https://github.com/markoblogo/abvx-agent-skills/tree/main/skills/diagnose
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install markoblogo-abvx-agent-skills@llmmart
Git git clone https://github.com/markoblogo/abvx-agent-skills.git

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

Skill manifest

Diagnose

The core job is to build a reliable feedback loop. Without one, hypotheses are weak.

Phase 1: Feedback Loop

Create the smallest useful pass/fail signal:

  • failing test;
  • CLI command with fixture;
  • HTTP request against local server;
  • browser script;
  • replayed trace or payload;
  • small harness around the relevant function;
  • repeated flake trigger with counted failure rate.

If no loop can be built, state what was tried and ask for logs, traces, access, or permission to instrument.

Phase 2: Reproduce

Confirm the loop matches the user's symptom, not a nearby failure. Capture exact error text, wrong output, timing, network failure, or UI state.

For nondeterministic issues, raise the reproduction rate with repetition, stress, seeded randomness, timing controls, and isolation.

Phase 3: Hypothesize

Write 3-5 ranked hypotheses before editing. Each must include:

Hypothesis:
Prediction:
Probe:
Result:
Status:

Test one variable at a time. Update rankings as evidence arrives.

Phase 4: Instrument

  • Prefer debugger, REPL, targeted logs, traces, or profiler over broad logging.
  • Tag temporary instrumentation with a unique marker such as [DEBUG-7b3c].
  • Remove all temporary instrumentation before finalizing.
  • For performance regressions, measure first and fix second.

Phase 5: Fix And Regression Test

If a correct seam exists, turn the minimized repro into a failing regression check before the fix. If no correct seam exists, document that as an architecture gap.

Apply the smallest fix supported by evidence, then rerun:

  • the regression check;
  • the original feedback loop;
  • relevant project checks.

When the task needs an auditable incident, device, CI/runtime, or regression record, hand the established reproducer and verification commands to bug-evidence-protocol. That protocol records proof; it does not replace this diagnostic loop.

Final Report

Include root cause, winning hypothesis, files changed, verification, debug cleanup, and residual risk.

Files (abvx-agent-skills)
  • agents
    • openai.yaml 218 B
      Loading…
  • SKILL.md 2.4 KB
    ---
    name: diagnose
    description: Debug coding failures through reproduction, ranked hypotheses, narrow fixes, and verification. Use when a bug, regression, flaky test, performance failure, or inconsistent behavior is broken, slow, throwing, or hard to reproduce and the task needs disciplined diagnosis instead of guessed fixes.
    license: MIT
    metadata:
      abvx_status: experimental
      abvx_origin: adapted
      abvx_eval_tier: fixture_checked
    ---
    
    # Diagnose
    
    The core job is to build a reliable feedback loop. Without one, hypotheses are weak.
    
    ## Phase 1: Feedback Loop
    
    Create the smallest useful pass/fail signal:
    
    - failing test;
    - CLI command with fixture;
    - HTTP request against local server;
    - browser script;
    - replayed trace or payload;
    - small harness around the relevant function;
    - repeated flake trigger with counted failure rate.
    
    If no loop can be built, state what was tried and ask for logs, traces, access, or permission to instrument.
    
    ## Phase 2: Reproduce
    
    Confirm the loop matches the user's symptom, not a nearby failure. Capture exact error text, wrong output, timing, network failure, or UI state.
    
    For nondeterministic issues, raise the reproduction rate with repetition, stress, seeded randomness, timing controls, and isolation.
    
    ## Phase 3: Hypothesize
    
    Write 3-5 ranked hypotheses before editing. Each must include:
    
    ```text
    Hypothesis:
    Prediction:
    Probe:
    Result:
    Status:
    ```
    
    Test one variable at a time. Update rankings as evidence arrives.
    
    ## Phase 4: Instrument
    
    - Prefer debugger, REPL, targeted logs, traces, or profiler over broad logging.
    - Tag temporary instrumentation with a unique marker such as `[DEBUG-7b3c]`.
    - Remove all temporary instrumentation before finalizing.
    - For performance regressions, measure first and fix second.
    
    ## Phase 5: Fix And Regression Test
    
    If a correct seam exists, turn the minimized repro into a failing regression check before the fix. If no correct seam exists, document that as an architecture gap.
    
    Apply the smallest fix supported by evidence, then rerun:
    
    - the regression check;
    - the original feedback loop;
    - relevant project checks.
    
    When the task needs an auditable incident, device, CI/runtime, or regression record, hand the established reproducer and verification commands to `bug-evidence-protocol`. That protocol records proof; it does not replace this diagnostic loop.
    
    ## Final Report
    
    Include root cause, winning hypothesis, files changed, verification, debug cleanup, and residual risk.
    
  • SKILL_CARD.md 1.1 KB
    Loading…

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.