TASK GUIDE · SOURCE-AWARE

Agent Skills for debugging and bug reproduction

A disciplined Skill plan for repository intake, minimal reproduction, and regression evidence.

Build this plan for my Agent ↗
Desired outcome

Reduce an intermittent report to the smallest repeatable failure and a regression check.

Safe working boundary

Observe first; change one variable at a time and preserve the original failure evidence.

01 · SKILL STACK

A minimal three-role stack

Each role has one job. The personalized builder can adjust the selection for your Agent without changing the decision logic.

01 · Core execution

repo-intake-and-plan

Maps the relevant execution path before changing code.

Main limitation
It does not create environments, download assets, execute commands, perform paper lookup, or write the final report.
First risk
Planning from incomplete repository evidence
Source evidence
Pinned · 3ab50525
Inspect the Skill ↗
02 · Guardrail

grill-me

Forces a minimal run and challenges unsupported assumptions.

Main limitation
A relentless interview can be tiring and disproportionate for small, reversible decisions.
First risk
Decision fatigue
Source evidence
Pinned · 885e2ca4
Inspect the Skill ↗
03 · Evidence

tdd

Turns the reproduced failure into a lasting regression check.

Main limitation
The workflow needs a functioning test runner and clear public behavior.
First risk
False confidence
Source evidence
Pinned · 885e2ca4
Inspect the Skill ↗

02 · ACCEPTANCE

How to know the task is finished

  1. Reproduce the same failure twice from a documented starting state.
  2. Identify the smallest input or state transition that triggers it.
  3. Add a check that fails before the fix and passes after it.

03 · STOP CONDITIONS

When to stop

  • The reproduction depends on unknown external state.
  • Multiple variables are changed before the failure is isolated.
  • Logs may contain secrets or personal data.

04 · METHOD

How this guide was produced

SkillSignal maps a defined task to three non-overlapping responsibilities, then selects from reviewed catalog metadata. Agent fit, source state, permissions, and published scores are evaluated with deterministic rules. No LLM writes or reorders the recommendation at request time.