learning_ai_common_plat/.github/prompts/workspace-coverage-remediate.prompt.md
2026-03-21 12:40:38 -07:00

2.3 KiB

name description argument-hint agent
workspace-coverage-remediate Take the top safe coverage actions from a workspace audit and implement them incrementally with validation and isolated commits. Scope or remediation plan, for example "top 3 safe fixes", "web only", "Local Memory GPT and FlowMonk", or "tests only no source changes" agent

Use the current workspace coverage findings to take the safest, highest-signal remediation actions across the current multi-repo workspace.

Goal

Select a small number of low-risk, high-value actions and complete them end to end. Default to the top 3 safest actions unless the user specifies a different scope.

Safety Rules

  • Respect repo boundaries.
  • Respect existing dirty worktrees and other in-progress changes.
  • Stage and commit only the files you changed.
  • Prefer isolated incremental commits, one repo or one logical fix at a time.
  • Avoid broad refactors.
  • Do not fix unrelated failures.

Selection Heuristics

Prefer targets that are:

  • pure logic modules
  • thin API clients or wrappers
  • hooks with deterministic behavior
  • small uncovered utilities
  • test-only additions when they add meaningful confidence

Avoid by default:

  • broad UI snapshot work
  • unstable E2E surfaces
  • setup-heavy mobile or native test changes
  • changes that require dependency installation unless explicitly approved

Execution Process

  1. Review the current coverage findings or infer them from nearby reports and tests.
  2. Choose the smallest safe target.
  3. Implement the change.
  4. Run targeted validation first, then repo-level validation if reasonable.
  5. Commit only your files with a conventional commit message.
  6. Repeat for the next target until the requested scope is complete.

Output Requirements

For each remediation step, report:

  • chosen target
  • why it was selected
  • files changed
  • validation run
  • commit hash
  • remaining risks or blockers

Default Mode

If no argument is provided:

  • take the top 3 safe actions
  • prefer web and shared-package targets first
  • keep each change small and independently commit-ready

Constraints

  • If a repository is dirty, work around unrelated changes rather than reverting them.
  • If a target conflicts with other active changes, skip it and choose the next safe candidate.
  • If no safe code change is available, produce a ranked remediation queue instead of forcing edits.