Software Development Skills

Apply organization-wide rules (from merge-rules output) to the current project. Detects tech stack, merges Principles, cleans up promoted patterns from .local.md, and fixes non-conforming files.

Asks Antigravity CLI for coding assistance. Use for getting a second opinion, code generation, debugging, or delegating coding tasks.

Manage macOS caffeinate to prevent system sleep. Supports start/stop/status. macOS only.

Monthly maintenance routine for dev-workflow's char-budget health (G2b — "reduce what already exists", the counterpart to G2a's "stop growing" cap and A1's "reduce via structure"). Consolidates duplicate reminders in skills/dev-workflow/SKILL.md / references/.md, re-audits intentional reinforcement-by-repetition, converts synonymous prose to cross-references, re-opens dev-workflow-triage issues that were deferred by the char-budget gate, dispatches Skill(verify-skill-refs) for governed-site enumeration drift, and checks G4 history-note sunset eligibility. Project-local routine — not for marketplace distribution. Trigger manually (/dev-workflow-monthly-consolidation) on a monthly cadence, or sooner when dev-workflow-triage's summary reports a growing budget-deferred (skipped, awaiting headroom) count.

Triage open issues in the dev-workflow-bundle retrospective repo. Read each open issue, judge each Finding (accept / reject), apply accepted fixes to the triage-scope skills (the bundle skills dev-workflow, ask-peer, extract-rules, rules-review, tidy, mobpro, plus dev-workflow-triage itself for self-targeted findings from its own self-retrospective), post a triage comment, and close the issue; after the per-issue loop, run a once-per-run rules-compliance detection pass (Step 3.8) over the run's diff via Skill(rules-review). Designed for non-interactive routine execution (no plan mode, no user prompts) on Claude Code on the Web.

Extract project-specific coding rules and domain knowledge from existing codebase, generating markdown documentation for AI agents. Use when onboarding a new project, after code review discussions about coding style, or when coding conventions need documenting. Also consider running after sessions where coding preferences were discussed or corrected (--from-conversation), or after PRs with significant review feedback (--from-pr).

Extract a named SKILL.md section's body verbatim into a references/ .md file, replace the body with a skeleton (section label + delegation pointer, runtime definitions left in place), sync the bundle copy, and verify byte-identity plus cross-reference integrity via verify-skill-refs / verify-bundle-sync. One-command automation of the manual section-extraction procedure used to shrink oversized SKILL.md files. Non-interactive — no user prompts. Project-local routine — not for marketplace distribution.

Read .claude/inbox-add.local.md and append one task line to the Obsidian Vault's inbox.md. Use when the user requests "add to inbox", "inboxに追加", "これをinboxに入れて", or any equivalent task-capture intent.

Merge extract-rules output from multiple projects into a unified portable rule set. Promotes .local.md patterns shared across projects to Principles format.

Learning-oriented development workflow (mob-programming style) that runs the SAME quality gates as dev-workflow — plan review, checks/tests, rules-compliance review, code review, interactive commits, rule maintenance — while pausing after every implementation unit for a diff review, so a junior engineer can follow WHAT is being built and WHY. The AI is always the driver (it writes all code) and narrates what it is doing and why; the junior navigates, reads each diff, asks questions, and approves commits. Use this instead of dev-workflow when the goal is to develop a feature WHILE a junior learns from it, not merely to ship efficiently.

Methodology for iteratively improving agent-facing instructions (skills / slash commands / CLAUDE.md / code-gen prompts) by having a bias-free executor run them and evaluating two-sidedly (executor self-report + instruction-side metrics) until improvements plateau. Use after creating or revising a prompt or skill.

Refactor verbose or unnatural natural-language prose — code comments, test descriptions, docstrings, user-facing text — into concise, native-sounding prose in a configured target language, using a sonnet subagent by default. Two modes: file mode rewrites a file's target-language prose in place; text mode returns the refactored text. Preserves code, identifiers, and proper-noun terms while translating ordinary technical vocabulary into the target language. Non-interactive — no user prompts. Use after generating prose with a model prone to verbosity, or to polish text before presenting it.

Review uncommitted diff for content unsuitable for publication to a public repository — secrets/credentials, user-specific absolute paths, internal-only URLs/hostnames, and personal identifiers. Each iteration dispatches a fresh subagent that returns findings and the main thread applies the subagent's mechanical fixes — a single pass by default, or re-dispatching up to Max iterations when the caller raises it. Non-interactive — no user prompts. Use as a final gate before publishing changes; designed to be called from non-interactive routines such as dev-workflow's hooks.oncomplete or dev-workflow-triage's per-Finding sub-flow.

Temporarily visualize untracked new files into the git diff so a chain of diff-deriving review callees can see them, then restore the files to untracked. Runs git add -N on untracked new files (intent-to-add), dispatches the requested review callees (rules-review / skill-review / publicity-review / verify-diff / tidy) in order, then git reset -- to restore the untracked state. Holds a closed list of each callee's visualization and restore-safety assumptions. Non-interactive — no user prompts. Project-local routine — not for marketplace distribution.

Check code changes for .claude/rules/ compliance. Use this skill when you need to verify that code changes follow project coding rules, whether as part of dev-workflow or standalone. Triggers on: rule compliance check, rules review, verify conventions, check coding standards. Best suited for hard rules (naming, imports, placement, explicit prohibitions); intent-style rules are checked on a best-effort basis.

Daily review of the latest origin triage- branch. Operator-prepared invariant — the operator fetches origin and switches the local repository to the triage branch before invocation. The skill verifies that the current branch matches triage-, then dispatches Skill(prompt-tuning) per prompt-eligible changed file, Skill(skill-review) with Base ref: main, Skill(publicity-review) with Base ref: main, and Skill(rules-review) with --base-commit main (rules-compliance detection layer) in sequence, finally emits a summary. Project-local routine — not for marketplace distribution.

Verify the dev-workflow-bundle canonical and bundle copy directories are in sync. Workaround for upstream symlink bug; delete this skill and its callers when symlinks are restored.

Empirically verify that a code diff achieves its stated objective by dispatching a bias-free executor that actually runs auto-derived evaluation scenarios against the post-diff file. The executor returns a JSON verdict with suggestededits, and this skill applies them iteratively until the executor declares no further fixes are needed, max-iterations is reached, or a safety rail trips. Non-interactive — no user prompts. Use after applying an Edit when you need a dynamic cross-check that complements static reviewers like skill-review.

Statically lint the dev-workflow and mobpro skill trees for SKILL.md cross-reference and output-naming invariants — dangling § / bold-label references, mirrored closed-list divergence, bare-number step references in prose, bare identifiers in user-facing output literals, and governed-site enumeration gaps — via a deterministic Node script plus a judgment subagent, returning a structured two-layer verdict whose pass/fail fields repeat identically across runs. Use after editing either tree's SKILL.md or its references/.md to catch cross-reference and naming drift before commit. Non-interactive — no user prompts. Project-local routine — not for marketplace distribution.