Perform grounded, broad code reviews that avoid tunnel vision and low-value findings. Use when asked to review code, changes, diffs, staged/unstaged work, branches, pull requests, or architecture impact of a change.
Ground agent hypotheses in verifiable evidence from source code, docs, issues, and concrete citations before acting. Use when investigating bugs, library behavior, architecture questions, ambiguous errors, or any task where guessing, assumptions, or stale knowledge could lead to wrong changes.
Make Jira Cloud REST API requests via a local helper script using credentials from environment variables or macOS Keychain. Use by default for Jira work: reading, searching, updating tickets, JQL, comments, transitions, ADF payloads, or any request involving Jira issues.
Create or Update Pull Request on Github
Save important information from the current agent conversation into a human-readable Markdown document. Use when the user invokes /save-to-md, asks to save an investigation/decision/summary to an .md file, or wants conversation context captured as documentation or an ADR-like record.
Investigate bugs/issues to find concrete, evidence-backed reproduction steps without fixing them. Use when the user reports a bug, failure, regression, flaky behavior, stack trace, unexpected output, or asks to debug/reproduce an issue.
Use jj (Jujutsu) for local version control instead of git. Activate when: the repo has a .jj/ directory, the user or project config mentions jj, the user says 'use jj', or any version control operation is needed in a jj-managed repo. Also use this skill when the user asks to commit, branch, stash, rebase, or perform any git-like operation in a repo that uses jj. If unsure whether the repo uses jj, check for a .jj/ directory.