Agent setup and playbook

Install the bundled skill #

rogrep skill install
rogrep skill show
rogrep skill uninstall

Installation always targets ~/.agents/skills/rogrep/SKILL.md. When ~/.claude, ~/.codex, or ~/.grok exists, rogrep also installs in that agent’s skills/rogrep/ directory. Read the rendered skill or raw source.

AGENTS.md snippet #

## Local conversation history (rogrep)
Use rogrep before debugging or rebuilding work a previous coding-agent session may already have investigated. Start with `rogrep QUERY` or `rogrep x QUERY`; narrow with `rogrep find CONVERSATION_ID QUERY`, then inspect with `rogrep show`, `rogrep git`, or `rogrep trajectory --pr N`. Use `--json` for automation and text for exploration. Cite conversation IDs, exchange references, and turn indexes. Do not launch `rogrep tui` from a non-interactive session.

Search, then drill down #

Recover context: rogrep x "incremental parser".
Locate exact turns: rogrep find rg_& "checkpoint".
Read the exchange: rogrep show rg&#e4 or rogrep show --around 27 rg&.
For a PR or branch, run rogrep trajectory --pr 48 and inspect candidates with rogrep git CONVERSATION_ID.

For failures, start with rogrep tool_status:failed tool_cmd:cargo. For file history, start with rogrep file:src/lib.rs. A tool failure without its surrounding request and repair attempt is incomplete evidence.

Output, evidence, and limits #

Use text while exploring and --json for automation. Cite rg& IDs, rg&#eN exchanges, and turn indexes. Prefer captured Git operations over text mentions when claiming causality.

An empty timeline is not proof nothing happened. Metadata or source records may be incomplete. Default corpus search excludes invisible harness turns and auxiliary conversations; conversation-scoped find searches everything. State those limits. rogrep tui requires an interactive terminal.