Retro reads the primary sources of a coding session, by default the current one, and looks for ways to improve the environment the agent works in. It checks seven areas: navigation, automated checks, coding standards, global AGENTS.md, tool economy, no-ops and information access. Candidates come back ordered by severity.
The skill is user-invoked and ends with a presentation of candidates for you to choose from.
What You Get
- A ranked list of environment improvements drawn from a real session
- Findings sorted into navigation, automated checks, coding standards, steering files, tool economy, no-ops and information access
- A check of the repo’s own lint and CI setup before any new guardrail is proposed
Common Use Cases
- Review a long session where the agent spent a lot of time finding the right files
- Decide whether a repeated mistake needs a lint rule, a hook or a CI job
- Trim an oversized CLAUDE.md or AGENTS.md after a session
What Makes It Unique
Retro aims the review at the environment the agent runs in. It separates mechanical violations, which get a deterministic check, from judgement calls, which go in CODING_STANDARDS.md. It also reasons about implementation and review as two stages with different context pressure, and gives the reviewer agent the job of enforcing standards.
Under The Hood
The skill is a single SKILL.md of instructions with no scripts. It first loads the writing-for-agents skill for style, then reads session logs on the local machine and the repo’s own check commands and CI workflow. It writes no files and runs no commands of its own; it returns a list to the user.