Agent-Skills.md

Agent Skills: Coding Agent Output Styles

Find, extract, and port coding-agent output style prompts across Claude Code, Pi, and OMP.

[Coding-AgentClaudePiOMPPromptingOutput-Style]
UncategorizedID: PoorRican/dotfiles/coding-agent-output-styles

Repository

PoorRican/dotfiles
PoorRican

Install this agent skill to your local

pnpm dlx add-skill https://github.com/PoorRican/dotfiles/tree/HEAD/configs/agent-skills/default/skills/autonomous-ai-agents/coding-agent-output-styles

Skill Files

Browse the full folder contents for coding-agent-output-styles.

Download Skill

Loading file tree…

configs/agent-skills/default/skills/autonomous-ai-agents/coding-agent-output-styles/SKILL.md

Skill Metadata

Name
coding-agent-output-styles
Description
"Find, extract, and port coding-agent output style prompts across Claude Code, Pi, and OMP."

Coding Agent Output Styles

Use this skill when asked to locate built-in or plugin-provided output style prompts, compare them, or port an output style from one coding agent to another.

Workflow

  1. Identify the source agent and installed version before searching.
    • Claude Code: command -v claude && claude --version
    • Pi: command -v pi && pi --help
    • OMP: command -v omp && omp --help
  2. Search readable config/plugin caches before reverse-engineering binaries.
    • Claude Code official plugin cache is often the best source for old/deprecated output styles.
    • Binary strings can confirm native remnants but should not be treated as the primary readable source when official plugin files exist.
  3. Extract exact prompt text from the source file and report the path alongside the text.
  4. When porting, prefer additive prompt injection over replacing the base system prompt.
  5. Avoid stacking styles that already include each other; this duplicates behavioral markers and makes the agent noisy.

Agent-specific injection targets

Claude Code

  • One run: claude --append-system-prompt <text> or claude --append-system-prompt-file <path>.
  • Persistent/project context: use Claude's normal CLAUDE.md, .claude/settings*.json, plugin, or custom command mechanisms depending on the scope.

Pi (pi, ~/.pi/agent)

  • Persistent additive behavior: ~/.pi/agent/APPEND_SYSTEM.md.
  • Project additive behavior: .pi/APPEND_SYSTEM.md.
  • One run: pi --append-system-prompt <file-or-text>.
  • Toggleable behavior: use a local extension under ~/.pi/agent/extensions/ that registers /output-style and appends the selected prompt in before_agent_start. Use templates/pi-omp-output-style-extension.ts as the starter and change the import to @earendil-works/pi-coding-agent plus AGENT_DIR to .pi/agent.
  • Avoid SYSTEM.md unless the task explicitly requires replacing/customizing the base prompt.

Oh My Pi (omp, ~/.omp/agent)

  • One run: omp --append-system-prompt <file-or-text>.
  • Persistent always-on behavior: prefer ~/.omp/agent/RULES.md; OMP treats top-level RULES.md as an always-apply rule reinjected near current turns.
  • Toggleable behavior: use a local extension under ~/.omp/agent/extensions/ that registers /output-style and appends the selected prompt in before_agent_start. Use templates/pi-omp-output-style-extension.ts as the starter.
  • ~/.omp/agent/SYSTEM.md exists, but is stronger and should be reserved for intentional core prompt customization.

Dotfiles/Home Manager pattern

For this user's dotfiles repo, prefer keeping Pi/OMP output-style artifacts in ~/dotfiles/configs/{pi,omp}/agent/ and exposing them through Home Manager home.file out-of-store symlinks. The useful shape is:

  • configs/pi/agent/extensions/claude-output-styles.ts
  • configs/pi/agent/output-styles/{explanatory,learning}.md
  • configs/pi/agent/output-style.json
  • configs/omp/agent/extensions/claude-output-styles.ts
  • configs/omp/agent/output-styles/{explanatory,learning}.md
  • configs/omp/agent/output-style.json

Wire these with a small Home Manager module (for example programs.coding-agent-output-styles) that uses config.lib.file.mkOutOfStoreSymlink and force = true; import it from nix/profiles/dev-extra.nix with lib.mkDefault true. If the user wants the change live immediately, replace the existing files with symlinks to the dotfiles copies before waiting for home-manager switch.

Pitfalls

  • Do not add Claude's Explanatory and Learning prompts simultaneously. Learning already includes Explanatory behavior, so stacking both duplicates the ★ Insight instruction.
  • Do not claim a style is unavailable just because it is not in the CLI help. Search plugin caches, settings/config directories, and install roots.
  • Do not rely on binary extraction as the authoritative prompt if an official plugin or documented file provides readable prompt text.
  • Pi/OMP extension hook APIs may have changed since examples mentioning systemPromptAppend: Pi 0.80.3 expects before_agent_start handlers to return a full systemPrompt string (event.systemPrompt + extra), while OMP 16.3.x expects a systemPrompt block array ([...event.systemPrompt, extra]). Preserve additive semantics even when the literal return field is no longer systemPromptAppend.

References

  • See references/claude-output-style-prompts.md for extracted Claude Code Explanatory/Learning prompt text and observed local paths.