Audit-and-merge editor for AGENTS.md files. AUTO-INVOKE this whenever a human asks to edit, modify, update, add, change, or remove a rule in an AGENTS.md file. It audits the ENTIRE target file, finds existing rules on the same topic, and MERGES the new rule into them instead of duplicating it. Contradictions, conflicts of interest, or any change that weakens a safety-critical rule are a HARD STOP for human decision. Never auto-commits: present the merged result as an unstaged edit plus a before/after diff. Use when changing AGENTS.md; reading AGENTS.md for context does not invoke it.
Commit code to git with a standardized provenance/approval trailer block — a single comma-separated Co-authored-by line with one entry per distinct author model (a single commit can mix authors, e.g. a generator wrote the impl and the adversary wrote the tests), Approved-by the local human's git identity, and a single Reviewed-by line listing the human approver first followed by any agent reviewer model IDs. Commits only what the human has already staged (never stages anything itself; if nothing is staged, nothing is committed); never pushes. Use when a human asks to commit reviewed changes and wants authorship/review provenance recorded in the commit message.
Create, list, and remove git worktrees for running multiple agents in parallel on the same repository without conflicts. Wraps fetch, sibling-path naming, submodule init, branch cleanup, and relinking gitignored local agent context back to the main worktree so it has a single source of truth. Use when the user asks to create, inspect, or clean up a git worktree for parallel branch or agent work.
Draft and publish reviewed GitHub issues with repository-aware gates. Resolve an explicitly named repo or the current checkout's remote, and ask when the target is ambiguous. Inspect repository issue templates and contribution guidance before drafting, run a mandatory duplicate search, gather adaptive evidence sections, validate requested or clearly implied metadata, persist markdown plus metadata in a separate git-backed workspace, and use the official GitHub MCP search_issues, issue_read, and issue_write tools with explicit auth failures or tool errors. Keep prose concise without duplicating the writer skill. Never invent identity. Use when preparing, reviewing, or filing a GitHub issue.
Author, validate and repair Agent Skills so they actually load. AUTO-INVOKE whenever creating, editing or debugging a skill or SKILL.md file, when porting skill scripts to Python, or when a skill that exists on disk never appears in an agent's context. Malformed skills fail silently: an over-long description, a name that does not match its directory, a BOM or unparseable YAML frontmatter each make the runtime skip the skill with no error in any log. Scaffolds new skills from a compliant template, lints frontmatter against the agentskills.io specification — load-blocking errors kept separate from conventions — and drives description-compression repairs that preserve trigger coverage. Enforces Python as the only source language and the shared _lib/skillkit package, flagging leftover shell scripts, missing shebangs or exec bits and non-portable shell-outs. Use when writing or fixing a SKILL.md; for AGENTS.md use agents-md.
Read comments and human-owned review marks in tuicr code-review sessions, open review windows in tmux, and run a live feedback loop. Includes a daemon that wakes an agent when comments are pending or a review closes, plus contracts for answering every comment before acting and interpreting deterministic approval without setting marks. Use when opening a review for a human, reading or replying to review comments, checking review completion, starting or stopping a live watch, or when woken by a tuicr wake or tuicr close prompt.
Distill human-facing documentation to the fewest clear words without changing its meaning, requirements, decisions, warnings, or tone. AUTO-INVOKE whenever drafting or revising documentation for human review, even when concision is not requested explicitly. Remove repetition, tangents, and unnecessary history or implementation detail; keep one coherent theme. Ask before omitting uncertain details. Use when writing or editing documentation a human will review.