Your coding agent has patience. This skill doesn't.
Skill Issue pushes AI coding agents toward smaller fixes, less unnecessary architecture, fewer repeated explanations, and more practical follow-through.
The attitude is optional. The competence is not.
Simple > clever. Boring > novel. Working > impressive.
npx skills add AccurateTLM13/skill-issueGlobal install:
npx skills add AccurateTLM13/skill-issue -gTry it without permanently installing:
npx skills use AccurateTLM13/skill-issue --skill skill-issueManual options: use SKILL.md directly, or download dist/skill-issue.skill for compatible harnesses.
Skill Issue biases an agent toward:
- the smallest real fix instead of the cleverest architecture
- working code before theory
- cutting scope creep and unnecessary dependencies
- recognizing repeated questions without inventing new theories
- asking fewer blocking questions
- explaining the failure without turning every bug into a lecture
- light roasting of the situation, never the person
Correctness outranks the bit. A shorter answer that is wrong is still wrong.
The repo includes a small A/B benchmark with real fixtures and objective verifiers. Executable cases must work before simplicity counts as a win; behavioral cases test restraint, tone, repetition handling, and safety.
Generated runs are created in a sibling skill-issue-runs/ directory outside this repo so the baseline agent cannot accidentally discover the root SKILL.md. The create script prints the exact path to use.
Example with S03:
node benchmark/scripts/create-run.mjs s03-dependency-hell baseline
# run the baseline agent against the printed directory
node benchmark/scripts/verify-run.mjs s03-dependency-hell baseline
node benchmark/scripts/measure-run.mjs s03-dependency-hell baseline
node benchmark/scripts/create-run.mjs s03-dependency-hell skill-issue
# repeat with the same model/settings and Skill Issue enabled
node benchmark/scripts/verify-run.mjs s03-dependency-hell skill-issue
node benchmark/scripts/measure-run.mjs s03-dependency-hell skill-issueThe measurement report shows files changed, line-count change, dependencies before/after/added, and introduced build/config files.
Read the benchmark methodology →
SKILL.md is the canonical Agent Skills entrypoint and includes name, activation-focused description, license: MIT, author metadata, and version metadata.
MIT licensed. See LICENSE.
v0.1.0 is actively being tested across models and agent harnesses. Results where the baseline wins are just as useful as results where Skill Issue wins.