Menu
SkillsMP has collected 18 skills from tomzhengy/harness-configs. Open a skill to review its source and details.
Showing 18 of 18 collected skills.
Delegate implementation, migrations, tests, experiments, investigation, and data analysis directly to a gpt-5.6-sol subagent. This is the default implementation path when taste-sensitive design is not the primary constraint.
Ask a direct gpt-5.6-sol subagent for an independent review of uncommitted changes, a branch diff, a commit, or a specific implementation. Use when a second review perspective is useful.
Delegate tasks that need a live browser or gui interaction (click-through verification, web flows, visual checks) to codex with computer use enabled.
Fetch the current clipboard image from the local mac via clipaste and read it. Use when the user says /clip, "check my clipboard", "look at my clipboard", or "what did i copy".
GitHub PR/issue agent transcripts: redact, preview, and insert safely.
GitHub PR/issue agent transcripts: redact, preview, and insert safely.
Continuously babysit the user's open GitHub PRs from a dedicated loop session. Run /babysit in a fresh chat to start a perpetual loop - each cycle finds the user's open authored PRs, then for each one fixes failing CI, syncs/resolves base conflicts, addresses…
Enable away mode so Claude can place a ringing phone call when it hits a blocker while you are asleep or away. This is triggered only by you typing /away; Claude never invokes it on its own.
Place a literal ringing phone call to alert the user only when Claude is blocked and cannot make progress. Use ONLY when (1) Claude is blocked and needs the user's decision or input to proceed, or (2) something failed unrecoverably and work cannot continue.…
Turn off away mode because you are back. Triggered only by you typing /back; Claude never invokes it on its own.
Run an extremely strict maintainability review for abstraction quality, giant files, and spaghetti-condition growth. Use only when the user asks for a thermo-nuclear code quality review, thermonuclear review, deep code quality audit, or especially harsh…
Claude Code second opinion from inside Codex. Three modes: review (diff review with a pass or fail gate), challenge (adversarial failure hunting), and consult (ask questions with locally prepared context). Use when asked for `/claude`, "claude review",…
Simplify code after changes are made without changing behavior or public APIs. Use when the user asks for simplification, when a recent change is correct but overly complex, or when a targeted readability pass is needed after implementation.
Review code changes for bugs, regressions, security issues, and missing tests without editing files. Use when the user asks for a review, before a PR, after a refactor, or when a recent change needs a focused quality pass.
Merge the current git worktree branch back into a target branch and clean up the worktree. Use when the user wants to finish a worktree-based task, merge the worktree branch into `main` or another base branch, and remove the temporary worktree safely.
Draft concise conventional commit messages from the current git changes. Use when the user asks for a commit message, wants help summarizing staged or unstaged diffs, or needs a one-line commit title that matches the repo's commit style.
Use Nia as the primary external research system for remote repositories, documentation, packages, and research handoffs. Trigger this skill when Codex needs grounded external information before implementation, when a user asks to research or verify something,…
Research a codebase and produce decision-complete implementation plans before code is written. Use when the user asks for a plan, wants to compare implementation approaches, or needs a concrete execution strategy with impacted files, risks, and tests.