Menu
Commit every dirty file in the current repo, only when on main.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Run an HQ-issued Slack bot — arm a per-bot @-mention watcher that spawns an autonomous Claude worker per mention. Pass the bot's slug (e.g. `hassaan`), a scope flag (`-c <company-slug>` or `--personal`), and optionally a workspace (`-w <slack-team-domain>`; auto-derived for `--personal` from SLACK_CREDENTIALS_JSON). The watcher resolves scope/workspace before personUid, using the sole local cache entry or an exact selected-vault bot-token namespace match; pass `-u <prs_…>` to override. It verifies and injects scoped `ANTHROPIC_API_KEY` into detached workers, handles the one-time bypass-permissions gate, resolves `<personUid>/HQ_SLACK_BOT_TOKEN_<NAME>_<WORKSPACE>`, infers the creator's Slack user_id (used as a DM gate), enumerates channels, and dispatches workers. Workers respond in-thread, ignore DMs from non-creators, and never call AskUserQuestion.
Surface architectural friction and propose deepening opportunities — turn shallow modules into deep ones for better testability and AI-navigability. Output: ranked candidate list with deletion-test outcome, leverage/locality scoring, file refs. Never edits code directly; presents candidates and walks the user through grilling-style design decisions for picked candidates. Use when the codebase is hard to change, when /diagnose hands off "no good test seam," or when planning a refactor wave.
Shared vocabulary and discipline for designing deep modules (a lot of behaviour behind a small interface, at a clean seam). Use when designing module boundaries or interfaces, or when another skill needs the deep-module vocabulary. Triggers on "deep module", "interface design", "design it twice".
Disciplined diagnosis loop for hard bugs, performance regressions, and intermittent failures. Build a deterministic feedback loop FIRST, then reproduce → hypothesise (3-5 ranked) → instrument (one variable at a time, tagged probes) → fix at the correct seam → cleanup + post-mortem. Use when the dominant problem is "I cannot reliably reproduce or measure this." For bugs that reproduce reliably with unknown root cause, use /investigate instead.
Actively build and sharpen a project's domain model — challenge fuzzy or overloaded terms against a glossary, stress-test with edge cases, and maintain CONTEXT.md as the source of truth for a ubiquitous language. Use when the user wants to pin down domain terminology, resolve a naming conflict, define a ubiquitous language, or when another skill needs to produce or sharpen the domain model. Triggers on "what should we call this", "domain model", "ubiquitous language", "define these terms", "update CONTEXT.md".
Enforce test-driven development with RED→GREEN→REFACTOR cycle and coverage validation
Based on SOC occupation classification
| name | commit-main |
| description | Commit every dirty file in the current repo, only when on main. |
| allowed-tools | Bash(git:*), Bash(date:*), Read |
Codex adapter for /commit-main.
Arguments: <commit message>
Read .claude/commands/commit-main.md first. That slash command owns the workflow, branch guard, staging behavior, and completion report.
Execute the command workflow inline from the current repository.
git branch --show-current returns exactly main before staging.git status --short to detect whether there are changes to commit.git add -A because the command's purpose is to commit every dirty path, including deletions and untracked files.git commit -m "<commit message>".Report whether a commit was created. If the command stopped because the branch was not main, no commit message was provided, or the repo was already clean, say that directly.