Agent-Skills.md

Agent Skills: Write maintainable code

Evaluate minimum-sufficient implementation routes and enforce the selected route for a fixed, authorized outcome. Use after outcome, scope, authority, and acceptance evidence are fixed. Do not choose outcomes, set acceptance, adjudicate governance, grade finished work, design tests alone, conduct security audits, or perform unrelated cleanup.

UncategorizedID: Ezra144israel/governed-agent-skills/write-maintainable-code

Repository

Ezra144israel/governed-agent-skills
Ezra144israelLicense: MIT
202

Install this agent skill to your local

pnpm dlx add-skill https://github.com/Ezra144israel/governed-agent-skills/tree/HEAD/skills/write-maintainable-code

Skill Files

Browse the full folder contents for write-maintainable-code.

Download Skill

Loading file tree…

skills/write-maintainable-code/SKILL.md

Skill Metadata

Name
write-maintainable-code
Description
"Evaluate minimum-sufficient implementation routes and enforce the selected route for a fixed, authorized outcome. Use after outcome, scope, authority, and acceptance evidence are fixed. Do not choose outcomes, set acceptance, adjudicate governance, grade finished work, design tests alone, conduct security audits, or perform unrelated cleanup."

Write maintainable code

Use the fewest concepts that satisfy the approved result and its acceptance evidence. Minimize maintenance burden, not raw line count.

Keep the ownership boundary

  • reasoning-doctrine owns the stage, outcome, scope, evidence, route recommendation, hard stops, and the transition to execution.
  • governed-operator owns authority, seats, formal contracts, verdicts, publication, and Done.
  • This skill owns route comparison, ownership seams, concept control, readability, and implementation verification.

If outcome, acceptance evidence, or authority is unresolved, return to its owner. Apply this lens only to the selected route and safe slice. A finished change needs independent review.

Compare routes before adding code

  1. Restate the fixed result and acceptance evidence without reopening them.
  2. Inspect current behavior, configuration, reuse, deletion, documentation, and no-code routes. If the current evidence already passes, make no change.
  3. Choose a route only when the fixed acceptance evidence can prove it.
  4. Name the smallest ownership seam and the concepts the selected route needs. Name tempting concepts the result does not need.

When routes satisfy different checks and the current evidence does not identify the applicable check, return that gap to whoever owns the outcome and acceptance evidence. Do not create a second implementation brief or return schema.

Bound the implementation

Decline a dependency, layer, abstraction, state, endpoint, migration, generic helper, or compatibility path whose only reason is hypothetical reuse or an unselected future. Prefer readable, explicit, testable code over dense code. A longer implementation is smaller when it carries fewer concepts or less maintenance risk.

Treat file size as a signal, not a diagnosis. Extract a bounded seam only when the extraction lowers verified current risk. Remove only code made obsolete by the selected change. Report unrelated cleanup instead of doing it.

For external calls, choose the failure path deliberately: timeout, retry, idempotency, partial failure, and secret handling where they apply. Load project security or infrastructure detail for those boundaries.

Verification and repair discipline

Before editing, inspect the repository status and the complete target diff. Separate defects introduced by the selected change from pre-existing defects. Pre-existing issues stay out of scope unless the operator expands the contract.

Match verification to the fixed acceptance evidence. Before returning, inspect the final diff and status again, then run the required repository gates. Advisory or judgment-heavy measurement must not become a mandatory deployment gate. After each repair, run the narrowest relevant focused recheck. Never weaken assertions, remove coverage, suppress errors, or change expected behavior to obtain a green result.

Stop when the selected acceptance evidence passes. Do not add cleanup or future-proofing after that point.

With governed-operator active, return the route, seam, declined concepts, verification, and residual risk in its Builder judgment record.