Draft or revise external communication in the user's concise, direct, informal voice. Use whenever Codex is asked to write, draft, reply, post, or comment on the user's behalf, including posting comments for them without an explicit "as the user" instruction. This skill controls wording only and never grants permission to send or post.
Use when working with a Backlog.md project or its `backlog` CLI/MCP, including creating, reading, refining, updating, completing, or archiving tasks and managing dependencies or milestones.
Refine selected backlog items or whole backlog sources into implementation-ready work. Use when asked to clarify, decompose, or prepare work across Markdown, Backlog.md, GitHub Issues, Linear, or another supported provider.
Resolve and operate on backlog sources consistently across loose Markdown, Backlog.md, GitHub Issues, Linear, and other supported providers. Use whenever a backlog command or skill reads or changes provider-backed work.
Resolve human decisions, missing inputs, stale blockers, and decision-only backlog items. Use when asked to unblock work or persist answers across Markdown, Backlog.md, GitHub Issues, Linear, or another supported provider.
Coordinate multiple local agents working from the same backlog or work queue with bounded Worklease claims. Use when selecting, claiming, heartbeating, checkpointing, handing off, or releasing shared work.
Explicit full implementation-review method for correctness, security, performance, maintainability, tests, and likely future failure modes. Use only when a full-review trigger applies to a branch, PR, commit range, changed files, or completed backlog item; do not use for light or style-only review or backlog state mutation.
Diagnose a bug, failing test, error, regression, or incorrect runtime behavior by reproducing it and tracing the first causal divergence. Use when the true source is unknown; do not use for speculative cleanup or when a proven fix is already specified.