harness

Plugin harness from marketplace agentic-harness in joe-yama/agentic-harness. Commands read from the repo's manifest — 1 month ago.

Skills (5)

• adopt

  Sets up the agentic-harness in a new or existing repository (Copier template, plugin, OpenSpec, branch ruleset) and verifies the guard hooks are live; also updates an adopted repository to a newer harness version. Use when asked to adopt, install, bootstrap or update the harness.

• design

  The design session of a change - read the context, choose the route, ask the PO one question at a time, then write the whole change (proposal, delta spec, design, tasks that double as the plan) with OpenSpec propose, create the one Issue and end the session. Use when the PO brings a new idea or a be

• execute

  The implementation session of a change - read the change, keep a ledger, dispatch harness:implementer per task, review in solo or batched units with harness:reviewer-intermediate, run fix rounds, rebuild when fixing fails, then one final review with harness:reviewer, one push and the PR. Use when st

• mutation-check

  Proves that a test is a real guardian by breaking the code it guards in an isolated copy of the repository and watching the test go red, with a control run. Use for refusal tests for authorization and data access, security tests, and tests the reviewer doubts; nothing else.

• workflow

  The per-change lifecycle of a PO-steered repository - design, OpenSpec proposal plus one GitHub Issue, change size budget, implementation with review, PR, archive and status update in the PR, where each record lives, one new session per stage, and the periodic retro on the harness's own cost. Use wh

Subagents (3)

• implementer

  Implements the task(s) of a change's tasks.md that the controller names, with strict TDD, and returns an evidence-backed report. Dispatch it with the model from your agent's section of docs/harness/models.md; never use it to review.

• reviewer-intermediate

  Reviews the commits of one task or batch against the parts of the change those tasks name, adversarially and in a context separate from the implementer - spec compliance with a scenario-to-test map, then code quality, then over-engineering - and returns a verdict. Use for task and batch reviews and

• reviewer

  Reviews a range of commits against a change's delta specs, design and tasks, adversarially and in a context separate from the implementer — spec compliance with a scenario-to-test map, then code quality, then over-engineering — and returns a verdict. Use for the final whole-branch review of a change

Hooks (7)

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/ask-gate.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh"

• PostToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/lint-on-edit.sh"

• SubagentStart

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/subagent-start.sh"

• SubagentStop

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/subagent-stop.sh"

• Stop

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/test-on-stop.sh"

Similar plugins