Test and debug

Skill olgaiv39/claude-oss-skills/skills/test-and-debug

Diagnose and repair a failing test, build, or runtime error in a public open-source project using the narrowest reproduction and validation. Use when something is broken and the cause is not yet known. Classifies the failure, isolates a reproduction, applies the smallest fix, and confirms with a targeted check. Does not add features.From its SKILL.md

Install

npx -y skills add olgaiv39/claude-oss-skills --skill test-and-debug

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

12 stars12 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.0 KB, ~1.0k tokens by cl100k_base, as published. Nobody here has run it

test-and-debug

Diagnose a failure, isolate the narrowest reproduction, apply the smallest fix, and confirm with a targeted check. Do not add features or refactor unrelated code while using this skill.

Activate when

A test, build, typecheck, or runtime path is failing
The cause of a failure is not yet known
A previously passing check now fails

Do not activate when

The change is a new feature with no failure yet -> use implement-minimal
No plan exists for a non-trivial change -> use oss-plan
The failure is a dependency install or version conflict only -> use dependency-review

Required inputs

The failing command or the observed error
Access to the repository to reproduce and inspect
Whether the failure is new or pre-existing, if known

Low-resource policy

Read the first of these that exists, then follow it:

${CLAUDE_PROJECT_DIR}/.claude/shared/LOW_RESOURCE.md
$HOME/.claude/shared/LOW_RESOURCE.md

If neither exists, apply this fallback: run one expensive command at a time, prefer the narrowest validation, disable watch mode, reuse existing environments, and run full validation only at a milestone boundary. Do not scan the whole filesystem to locate the policy.

Context-efficiency policy

Read the first of these that exists, then follow it:

${CLAUDE_PROJECT_DIR}/.claude/shared/CONTEXT_EFFICIENCY.md
$HOME/.claude/shared/CONTEXT_EFFICIENCY.md

If neither exists, apply this fallback: select files before reading; use targeted searches and bounded ranges; do not preload references; do not reread unchanged files; finish one atomic increment and stop; create a compact handoff before context is exhausted.

Facts that must not be assumed

The test runner, package manager, or build tool
That the failure is deterministic
That the failure is caused by the most recent change
That an external system the code calls is currently reachable

Preflight

git status --short and git diff --stat to see uncommitted work
git log --oneline -5 to see recent changes that may correlate
Discover the project's test and build commands -> references/test-discovery.md
Capture the exact failing command and its output

Workflow

Reproduce the failure with the narrowest command that triggers it -> references/test-discovery.md
Classify the failure -> references/failure-classification.md
Isolate: reduce to the single test, input, or code path that fails
Form one hypothesis about the cause and the smallest evidence to confirm it
Confirm the cause before editing; do not fix by guessing
Apply the smallest fix for that cause -> references/recovery-recipes.md
Re-run the narrow reproduction; confirm it passes
Run the related test file to check for a regression in the same area
If the fix changed public behavior, note the documentation impact
Inspect git diff and remove debugging scaffolding and stray edits
Produce the report using templates/debug-report.md
Stop after the fix is confirmed

Failure classification

Classify before fixing; the class determines the recovery -> references/failure-classification.md

Assertion failure (wrong result)
Error or exception (crash)
Build or compile failure
Type or lint failure
Flaky or nondeterministic failure
Environment or dependency failure
External-system or boundary failure
Timeout or resource-exhaustion failure

Recovery actions

Match the recovery to the class; apply the smallest one that resolves the confirmed cause -> references/recovery-recipes.md

Validation escalation

narrow reproduction command
single failing test
related test file
changed-file lint or typecheck
full suite only at the milestone that ends the work

Do not run the full suite to find which test fails; find it with the narrow command first.

Stop conditions

The failure cannot be reproduced locally after one focused attempt
The failure depends on an unavailable external system with no mock
The fix would require changing a public API or auth path without approval
The root cause remains unconfirmed after one hypothesis-and-evidence pass

Human review boundaries

A fix that touches auth, wallet, or user-data handling
A fix that changes a public API surface
A fix that can only be validated against an unavailable live system

Final report

Produce the report in the exact section order of templates/debug-report.md, then stop.