Use this skill whenever the user is deliberately starting or resuming work for an existing epic in its own worktree. This includes explicit requests like "start work on ABU-123", "create a worktree for this epic", "set up the epic worktree", "/cape:worktree", or being on the default branch before BUILD work. Do NOT use for ordinary git branch creation, PR preparation, rebasing, merging, deleting worktrees, or creating the epic itself.
Use BEFORE writing any code for new features, integrations, or system changes. Triggers when the user describes something to build, asks "how should I approach X", is unsure between approaches, or mentions adding/creating/building functionality. Also use when requirements are vague, architecture is unclear, or the task involves design decisions (e.g., choosing libraries, data models, API patterns). Do NOT use for bug fixes, straightforward code cleanup, executing existing plans, or tasks where the implementation path is already clear. This skill researches the codebase, asks Socratic questions, generates competing designs under different constraints, and produces a design summary for `cape:write-plan` to formalize into a Linear tracker epic.
Create atomic git commits with conventional commit format and selective staging. Use whenever the user wants to commit changes — explicit requests ("commit this", "make a commit", "let's commit", "/cape:commit") and implicit ones ("we're done, save this", "wrap this up"). Also use when another cape skill finishes a unit of work and needs to commit. Covers staging decisions, splitting large diffs into separate logical commits, and writing thorough commit messages that explain the change. Do NOT use for pushing, creating PRs, or branch operations.
Meta-skill that activates cape's workflow system. Injected at session start; always active and never manually triggered. Routes every task to the right cape skill and enforces workflow chains: brainstorm before planning, plan before coding, TDD during implementation, diagnosis before fixing. If a cape skill matches the user request, using it is mandatory.
Build from a Linear tracker epic, one task at a time. The counterpart to cape:write-plan: write-plan creates the epic and first sub-issue, execute-plan implements it. Triggers on: "continue", "next task", "resume", "let's go", "work on the plan", a Linear issue ID, or transitioning after planning is complete. Uses the local tracker cache for orientation and refreshes that cache after every Linear write. Do NOT use for bug fixes (use cape:fix-bug), exploratory design (use cape:brainstorm), or acting on inbound PR review comments (use cape:pr-feedback).
Verify and hand off a Linear tracker epic after all tasks are complete. Use when the user says "finish the epic", "we're done", "close out the epic", all tasks are done and the user wants to close it, or execute-plan detects all acceptance criteria appear met. Runs final verification, checks acceptance criteria with evidence, optionally writes a minimal outcome summary to Linear, and hands off to review/PR. Linear's GitHub integration closes the epic when the PR merges; cape never sets Linear status.
End-to-end bug fix workflow from discovery to closure. Triggers on: "fix this bug", "can you fix", user wants to address a diagnosed Linear bug issue, fixing a regression, patching broken behavior, or investigating a defect before fixing it. Do NOT use for feature work (use execute-plan) or cleanup without a specific defect.
Resolve inbound PR review comments end to end: fetch the threads with their node IDs, judge each comment's validity, fix the accepted ones, then reply, resolve the matching threads via GraphQL, and commit referencing the review. Use whenever the user wants to act on review feedback: "fetch comments on PR, are they valid?", "resolve threads that are fixed", "push and resolve threads we fixed", "resolve the comment threads that are fixed or ignored", "fix the valid issues and resolve", "/cape:pr-feedback". Do NOT use for reviewing your own pre-PR diff (use cape:review), fixing a single diagnosed defect with no PR thread (use cape:fix-bug), or creating a PR (use cape:pr).