agent-harness

Plugin agent-harness from marketplace boss-skills in bossjones/boss-skills. Commands read from the repo's manifest — 6 days ago.

Skills (21)

• add-review-comment

  Post a single inline review comment to a GitHub pull request. Use when you need to leave one comment on a specific line or line range of a PR via the GitHub API — supports single-line and multi-line anchors and one-click suggestion blocks.

• boss-cmux-team

  Spawn, orient, and drive a multi-agent terminal team in cmux from natural language. Use when a prompt asks you to boot a team of agents (lead + workers) as a cmux workspace, stand up a "full-stack team" / "agent fleet", orchestrate several terminal agents on one feature, or attach to and drive a tea

• boss-cmux

  Drive cmux — the terminal multiplexer / agent-surface control CLI — from natural language. Use this whenever a prompt asks you to open, inspect, prompt, read, or tear down cmux windows, workspaces, panes, surfaces, or agent sessions. Prefix orchestration prompts with /boss-cmux.

• boss-security-review

  Performs a security review / security audit / vulnerability review of code and writes a structured, severity-graded findings report (default path specs/security-review.md, overridable from the request). Reviews changed code by default; the request can override the target to a named path (e.g. "revie

• fetch-diff

  Fetch a GitHub PR diff with old/new line numbers and auto-generated-file masking for code review. Use when you need a pull request's diff annotated with line numbers to place inline review comments, or want the diff filtered to specific file globs.

• fetch-unresolved-comments

  Fetch unresolved PR review comments via the GitHub GraphQL API, filtering out already-resolved feedback. Use when you need only the open review threads on a pull request — for example before addressing reviewer feedback or summarizing what still needs attention.

• git-worktree-clean

  Clean up stale git worktrees with merged-branch detection and a disk-usage report. Use when feature worktrees under .claude/worktrees/ have accumulated and you want to reclaim space — auto-removes merged worktrees and reviews unmerged ones.

• git-worktree-remove

  Safely remove a single git worktree with branch cleanup and safety checks. Use when you are done with one feature worktree and want to delete it plus its local and remote branches, with protected-branch, uncommitted-change, and merge-status checks first.

• git-worktree-status

  Report the status of background verification jobs (tests, type check, build) in a git worktree. Use when you want a non-blocking PASS/FAIL/RUNNING/NOT_RUN summary of the checks launched by the git-worktree skill, run from inside a worktree directory.

• git-worktree

  Create an isolated git worktree under .claude/worktrees/<repo>-<name>/ for feature development without switching branches. Use when starting a new feature, fix, or experiment that needs its own working directory — handles repo-prefixed naming, base-ref selection, .worktreeinclude file copying, and l

• github-pr-review

  Use when reviewing GitHub pull requests with gh CLI - creates pending reviews with code suggestions, batches comments, and chooses appropriate event types (COMMENT/APPROVE/REQUEST_CHANGES)

• harness-doctor

  Inspect agent-harness environment readiness and runtime storage without changing files. Use when asked to diagnose agent-harness, check prerequisites, find residual harness logs or data, learn which plugin repository names the harness root, inspect stale logs, .claude/data, or a legacy repository-na

• planf3

  Writes and maintains implementation plans as self-contained HTML files with synced visual identity, optional AI-generated diagrams, and a full create/update/build lifecycle with phase checkboxes and append-only metadata — saved to specs/. Use as a richer alternative to /agent-harness:plan when the p

• pr-review

  Review a GitHub pull request and emit a schema-validated local review payload — inline comments plus an approve-or-comment decision. Use when asked to review a PR, audit a pull request's changes, or produce review comments without posting them.

• pyrefly-typing

  Adopt Pyrefly (Meta's Rust-based Python type checker) into a target `uv` Python project as a non-blocking, agent-driven typing feedback loop, alongside whatever type checker it already uses (mypy/pyright/basedpyright/ty) — never replacing it. Use when asked to "adopt pyrefly", "set up pyrefly typing

• release-notes-generator

  Generate release notes in 3 formats (CHANGELOG.md, PR body, Slack announcement) from git commits. Automatically categorizes changes and converts technical language to user-friendly messaging. Use when preparing a release, drafting a changelog or version notes, or writing a what's-new or ship announc

• setup-agent-harness

  Make a repo "agent-harness ready" by safely updating .gitignore and .claude/settings.local.json. Use when setting up or onboarding a repo to the agent-harness plugin, ensuring .gitignore covers the plugin-namespaced legacy runtime-artifact root and optionally configuring the statusLine and an output

• setup-second-brain

  Install and configure the "second brain" — the obsidian-wiki uv tool plus optional QMD semantic search. Use when setting up a second brain, installing obsidian-wiki, making a machine "second-brain ready", configuring an Obsidian knowledge vault for AI agents, or enabling/configuring QMD semantic sea

• stop-slop

  Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells.

• unicode-hygiene

  Scan skill, plugin, command, agent, and marketplace files for invisible or visually-spoofed Unicode — tag characters (U+E0000–U+E007F), bidirectional overrides (Trojan Source), zero-width / format characters, and mixed-script (homoglyph) install lines. Use before committing or publishing skills/plug

• worktree-doctor

  Analyze a repo and suggest a .worktreeinclude so worktrees inherit env, secret, and local-config files. Use when setting up worktrees in a new repo, before the first /git-worktree, or when a worktree is missing its .env / local settings — scans gitignored files and proposes patterns.

Slash commands (16)

• /agent-harness:all_tools

  List every tool available in the current system prompt as bullet-point TypeScript signatures, suffixed with each tool's purpose.

• /agent-harness:autobuild

  Implement a spec inside a git worktree, then commit, push, open a PR, and address review comments. Must run inside a linked git worktree.

• /agent-harness:build

  Implement the plan

• /agent-harness:cmux-did-spawn

  Orient yourself as orchestrator over a multi-agent team that was just scripted into a cmux window. Reads the spawn file, locates the team by its (stable) window UUID, rediscovers surface refs, and gets ready to drive the lead.

• /agent-harness:cmux-fresh

  Clear cmux's persisted session so the next launch opens with fresh/blank windows. Quits cmux first (so the clear sticks), backs up, then empties the saved window tree (and its -previous sibling).

• /agent-harness:cmux-spawn-team

  Boot a multi-agent team as a new workspace in the cmux window (reuse the open window; only create one if none exists) — a lead on the left half, workers in a grid on the right, each launched with its role and model from the team config. Drive cmux yourself in natural language, or use the scripted fa

• /agent-harness:commit-push-pr

  Stage all changes, create a conventional commit, push to remote, and open a pull request via gh CLI. Use whenever you want to commit and ship current changes with a proper PR.

• /agent-harness:debug-ci

  Debug failed GitHub Actions CI runs — diagnose, fix, validate locally, commit, push, and poll until the new run passes. Supports up to 3 retry cycles.

• /agent-harness:docs-tutorial

  Kick off tutorial / documentation generation by delegating to `/documentation-generation:doc-generate`, routing the work to the correct fully-qualified tutorial-engineer subagent (documentation-generation for docs, code-documentation for code). Defaults to a tutorial on the current branch's features

• /agent-harness:fix-gh-pr-comments

  Fetch unresolved PR review comments via gh CLI, evaluate each one, apply fixes locally, validate, push, reply and resolve each thread, and poll for new comments. Supports up to 3 outer cycles.

• /agent-harness:install_status_line

  Install the agent-harness status line into this project's .claude/settings.local.json (backed up, reversible).

• /agent-harness:plan

  Creates a concise engineering implementation plan based on user requirements and saves it to specs directory

• /agent-harness:plan_w_team

  Creates a concise engineering implementation plan based on user requirements and saves it to specs directory

• /agent-harness:prime

  Load context for a new agent session by analyzing codebase structure, documentation and README

• /agent-harness:question

  Answer questions about the project structure and documentation without coding

• /agent-harness:validate-unicode-hygiene

  Scan files for invisible or visually-spoofed Unicode (tag characters, bidi overrides, zero-width / format chars, homoglyph install lines) using scripts/validate-unicode-hygiene.py. Defaults to the repo target globs; pass paths to scan specific files.

Subagents (4)

• hello-world-agent

  Simple greeting agent, use proactively when greeting the user. If they say 'hi claude' or 'hi cc' or 'hi claude code' use this agent.

• llm-ai-agents-and-eng-research

  AI research specialist that proactively gathers latest news and developments in LLMs, AI agents, and engineering. Use for staying current with AI/ML innovations, finding actionable insights, and discovering new tools and techniques.

• meta-agent

  Generates a new, complete Claude Code sub-agent configuration file from a user's description. Use this to create new agents. Use this Proactively when the user asks you to create a new sub agent.

• work-completion-summary

  Proactively triggered when work is completed to provide concise audio summaries and suggest next steps. If they say 'tts' or 'tts summary' or 'audio summary' use this agent. When you prompt this agent, describe exactly what you want them to communicate to the user. Remember, this agent has no contex

Similar plugins