Harness

Plugin harness in wcxcw/harness-skills. Commands read from the repo's manifest — 13 days ago.

Skills (19)

• agent-harness

  Initialize and operate a project-level Agent Harness for software engineering work. Use when the user wants to create or update AGENTS.md, scaffold a minimal harness/ directory, define project context/tools/guardrails/gates, create run records, or combine bundled workflow skills into a repeatable Ha

• meta

  Maintain and improve this harness skill pack, including workflow skills and their tests.

• writing-skills

  Create or improve workflow skills with clear triggers, outputs, gates, and verification.

• workflows

  Route closed-loop Agent Harness work through pre-run grilling and run-level skills such as writing-specs, writing-plans, executing-plans, code quality review, verification, and finishing-run.

• backend-quality-review

  Review backend, API, service, worker, persistence, and integration changes for boundaries, logging, error handling, and operability.

• code-quality-review

  Review completed code changes for maintainability, structure, observability, comments, and local conventions before final verification.

• code-reviewer

  Review Harness run changes for correctness, regressions, risks, and verification gaps.

• context-budget

  Decide which project harness context to load for a task and prevent bulk-reading long context files or completed run history.

• executing-plans

  Execute one approved `plan.md` task at a time while recording evidence in `execution-log.md`.

• finishing-run

  Close an active harness run with `evaluation.md`, residual risk, and harness feedback.

• frontend-quality-review

  Review frontend changes for component boundaries, state/data separation, accessibility, responsive behavior, and browser verification.

• grilling

  Interrogate a plan, decision, or idea until the user and agent reach shared understanding. Use only when the user explicitly invokes `$grilling`, asks to be grilled, or uses another clear grill trigger phrase.

• receiving-code-review

  Process code review feedback rigorously, verify each finding against the codebase, apply valid fixes one at a time, and record resolution in the active run.

• requesting-code-review

  Review completed changes before final verification and record findings in `review.md`.

• systematic-debugging

  Diagnose and fix bugs by reproducing, isolating cause, fixing narrowly, and adding regression evidence.

• test-driven-development

  Use a test-first loop when an automated regression test is valuable for the changed behavior.

• verification-before-completion

  Verify required evidence before claiming completion and make skipped checks explicit.

• writing-plans

  Turn an accepted `spec.md` into a scoped, ordered, verifiable `plan.md`.

• writing-specs

  Turn confirmed shared understanding into a run-level executable `spec.md` with scope, decisions, acceptance criteria, verification, and blocking questions.

Hooks (2)

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-.}}/scripts/harness_hook.py" SessionStart

• Stop

  python3 "${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-.}}/scripts/harness_hook.py" Stop

Similar plugins