Skip to content

Commands

Pegg's slash commands are implemented as skills: type /name in the input and Pegg issues a loadskill tool call, so the skill's instructions arrive as a tool result (or run as a background subagent for context: fork skills). Four commands are built in — /init, /batch, /review, and /rule — and you can add your own with a SKILL.md file.

In the TUI, type / to open the picker and filter by name. The same token works in the CLI (pegg run "/init") and inside subagent task messages.

/init

Instructs Pegg to analyze a codebase and create or improve an AGENTS.md file. Use it on a repository that lacks agent guidance, or to refresh an outdated one.

The generated file follows a fixed structure:

  • Project Overview
  • Repository Structure
  • Build & Development Commands
  • Code Style & Conventions
  • Architecture Notes
  • Testing Strategy
  • Security & Compliance
  • Agent Guardrails
  • Extensibility Hooks
  • Further Reading

Rules the init command follows:

  • The file must start with the # AGENTS.md heading.
  • Content is limited to roughly 12k tokens; keep it dense and factual.
  • Unresolved details are marked inline with > TODO: so you can fill them in.
  • Information the codebase doesn't reveal is not invented — it is left as a placeholder or omitted.

The result is a file the agent framework reads on every run as Project Instructions.

/batch

Instructs Pegg to orchestrate a large, parallelizable change from exploration to report, using subagents. This is the orchestrator's working contract for multi-file work.

The workflow:

  1. Decompose the task into independent chunks that can be parallelized.
  2. Explore the codebase to map the affected areas.
  3. Plan by writing a plan file under .pegg/plans/YYYY-MM-DD-<feature>.md.
  4. Delegate chunks to explorer, planner, and developer subagents via the task tool, running them concurrently.
  5. Verify the results before accepting them.
  6. Report the outcome, including what changed and what remains.

Supporting rules the batch command applies:

  • Subagent names are unique and task-descriptive; results can be re-used from the session's subagent store (.pegg/subagents/<session-id>.json).
  • The plan file is the source of truth for what is being changed and why.
  • Todo items use a hierarchy (todo-1, todo-1.1, ...), priorities, and dependsOn links so progress and dependencies stay visible.
  • Every subagent either proves its work (tests, checks) or reports a failure.

The lifecycle tools — task and taskstatus — back the whole flow. See Subagents.

/review

Instructs Pegg to review a pull request or changeset and produce structured feedback. Use it when you want a second pass over a diff before merging.

The review pipeline:

  1. Assess size — determine whether the review can be done in one pass or must be partitioned.
  2. Gather context — an Explorer subagent collects relevant background from the codebase.
  3. Review hunks — Developer subagents review the diff piece by piece, recording findings as:
- **[File:Line]** [Severity] - [Issue] - [Fix]
  1. Verify findings — a Verifier pass filters the list into [Confirmed], [Plausible], and [Refuted], dropping noise and duplicates.
  2. Sweep for gaps — look for issues the hunk-by-hunk review might have missed (security, tests, docs).
  3. Report — produce inline review comments on the VCS, or a Markdown report.

/rule

Creates or updates a rule file under <project>/.pegg/rules/ (project scope) or ~/.pegg/rules/ (global scope). See Rules for the full workflow.

Other commands

These are the only literal slash commands in the chat loop:

Input Effect
exit, quit, /exit, /quit, bye End the CLI run --chat loop

Anything else starting with / that is not a registered skill is left in the message text. There is no hidden command palette — Settings (Ctrl+P in the TUI) and the / picker are the full command surface.