harness-toolkit

Run PR-sized Harness Kit dogfood replay trials in temporary repos to validate HK lifecycle UX with realistic tasks and observe how agents naturally discover or misuse HK.

Mine repository validation contracts and propose or audit Harness Kit hk profiles. Use when choosing an hk profile, when no exact repo/module profile exists, when a user asks to create or update a custom profile, or when inspecting CI, hooks, mise, package scripts, repo docs, or HK config diagnostics to draft profile checks/reviews.

Router skill for Harness Kit config authoring and maintenance. Use when a user asks to create, audit, update, or repair HK profiles, system maps, target bindings, or dots/user-level harness.toml config. Delegates to profile and system-map authoring skills instead of duplicating their logic.

This skill creates, updates, audits, and validates Harness Kit system maps, including repo-local .harness/system.toml files and target-level user/dots system_map overlays. Use when authoring agent-facing component/invariant maps, connecting touched paths to relevant HK profile check labels, dogfooding system-map candidates, or preparing Harness Kit CLI integration trials.

This skill finds and attaches Pi, Claude Code, and Codex session or review transcripts to active Harness Kit work using `hk artifact attach`. Use when the user asks to capture agent sessions, attach transcripts, preserve Codex/Claude/Pi review evidence, or inspect HK lifecycle artifacts.

Skill-first workflow for planning, implementing, reviewing, and handing off a meaningful repo slice. Use when starting a slice, rendering planner/implementer/ reviewer prompts with mise, deciding what evidence to keep, or evaluating prompt quality with holdout sample tasks.

Creates, updates, or evaluates AGENTS.md and CLAUDE.md files for a repository. This skill activates when users want to create onboarding docs, update agent context files, or evaluate documentation quality. Generates docs with YAML frontmatter indexes for machine-readable progressive disclosure, system invariants, redirect-pattern gotchas (DO/NOT/BECAUSE), and validated commands. Supports incremental updates via git-based staleness detection.

Bring the active plan up to date before running sync-check. Focuses on META, TODOs, learning log coverage, and required slice-local files.

Compatibility wrapper for the implementation phase of the slice workflow. Use when implementing an active slice while keeping validation and evidence current.

Compatibility wrapper for the planning phase of the slice workflow. Use when shaping a slice before coding or when asked to plan a task with the generated slice workflow.

Compatibility wrapper for the review phase of the slice workflow. Use when preparing a slice handoff or validating review evidence.

Promote slice-local decision notes into durable repo docs. Uses the active plan's decision_record to decide between no-op, decision-ledger, or ADR.