Skill quality depends on current context

Dynamic context is the current issue and repository state supplied to a skill: arguments, search results, relevant files, code excerpts, diff, tests, and the template for the output document.

A strong skill selects the minimum sufficient context rather than passing the whole repository, while keeping the result format repeatable.

Context sourceRole
Issue and argumentsCurrent need
Search resultsCandidates for related modules
Files and code excerptsEvidence of actual behavior
Diff and testsChange state and result evidence
TemplateStable output structure

Dynamic-context flow

  1. Accept the issue as input.
  2. Find the most relevant files.
  3. Collect confirmation from code and tests.
  4. Apply the specification template.
  5. Separate facts from assumptions.
  6. Create TASK_SPEC.md with open questions.
  7. Stop before implementation.
Project-specific candidate search
grep -R 'refund' support/ orders/

Files beside a skill

  • SKILL.md — procedure and contract;
  • templates/ — artifact formats;
  • examples/ — expected-result samples;
  • references/ — supporting material.
.claude/skills/issue-analysis/
├── SKILL.md
├── templates/
│   ├── TASK_SPEC.template.md
│   └── EVIDENCE_LOG.template.md
├── examples/
│   └── refund-sorting-bugfix.md
└── references/
    └── risk-notes.md

Pre-commit quality review

  1. Confirm that the skill names one scenario.
  2. Check unambiguous input and output.
  3. Separate facts from assumptions.
  4. Check minimal tools and permissions.
  5. Confirm that only needed context is loaded.
  6. Run the procedure on a representative case and inspect the result.

Canonical lesson source · JavaRush

Open JavaRush material