From agent-supagents
Use when authoring or building supagents sources under .agents/supagents/ — markdown that compiles to subagent files for Claude Code, Gemini CLI, GitHub Copilot, Cursor, OpenCode, and Kilo Code.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-supagents:use-agent-supagentsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Compile one markdown source into N target subagent files. The persona ("You are a code-investigation expert…") is shared; per-target frontmatter is **verbatim** in named blocks.
Compile one markdown source into N target subagent files. The persona ("You are a code-investigation expert…") is shared; per-target frontmatter is verbatim in named blocks.
supagents, the ~/.agents/supagents/ directory, or wants to keep one persona in sync across Claude Code / Gemini CLI / Copilot / Cursor / OpenCode / Kilo..agents/supagents/ (project) or ~/.agents/supagents/ (global).# Generated by supagents from <path> marker — that file is generated; edit the source, not the output.uv tool install supagents (or pipx install supagents). Requires Python ≥ 3.12.^[a-z0-9][a-z0-9_-]*\.md$ — lowercase letters, digits, -, _.project if ./.agents/supagents/ exists in cwd, else global. Override with -g / -p.A minimal source:
---
name: code_investigator
description: Use to explore unfamiliar codebases — trace symbols, explain features.
CLAUDE:
model: sonnet
tools: Read, Grep, Glob, Bash
GEMINI:
kind: local
tools: ["*"]
---
You are the code-investigation agent. …
Frontmatter is partitioned by case:
| Case | Where it goes |
|---|---|
lowercase keys at top | Shared frontmatter — copied to every output. |
UPPERCASE keys at top | Target sections (CLAUDE:, GEMINI:, COPILOT:, …). |
lowercase keys inside T: | T's frontmatter; overrides shared on collision. |
UPPERCASE keys inside T: | Directives — not emitted to the output. |
Only two directives:
OUTPUT — override the output file path for this target only.APPEND_BODY — markdown appended to the shared body, target-only.A target section's presence triggers an emit. Drop the COPILOT: block to stop generating Copilot files. There is no field-level conversion — what you write is what gets written.
Bundled targets (override paths in ~/.config/supagents/config.yaml):
| Target | Global output | Project output | Suffix |
|---|---|---|---|
CLAUDE | ~/.claude/agents/ | .claude/agents/ | .md |
GEMINI | ~/.gemini/agents/ | .gemini/agents/ | .md |
COPILOT | ~/.copilot/agents/ | .github/agents/ | .agent.md |
CURSOR | ~/.cursor/agents/ | .cursor/agents/ | .md |
OPENCODE | ~/.config/opencode/agents/ | .opencode/agents/ | .md |
KILO | ~/.config/kilo/agents/ | .kilo/agents/ | .md |
supagents init NAME [-g | -p] [--force]
supagents build [-g | -p] [--dry-run | --check] [--target T]... [--config PATH] [-v]
supagents clean [-g | -p] [--dry-run] [--target T]... [--config PATH]
supagents list [-g | -p] [--target T]... [--config PATH]
build is idempotent — unchanged outputs aren't rewritten (mtimes preserved).--check exits 1 if any output would change. Useful as a CI guard against stale generated files. Mutually exclusive with --dry-run.--target is repeatable, case-insensitive; unknown targets are rejected.clean deletes only files carrying the supagents generation marker — hand-written siblings are preserved.0 clean, 1 per-target errors or --check drift, 2 fatal parse errors.Add a new agent (project scope):
supagents init code_investigator --project # writes .agents/supagents/code_investigator.md with all six target blocks
$EDITOR .agents/supagents/code_investigator.md # tweak frontmatter, write the persona body
supagents build --project # emit per-target files
Drop a target without losing the source: delete the XXX: block from the source frontmatter and run supagents clean to remove the orphaned outputs.
CI guard: add supagents build --check to CI — fails if checked-in generated files are stale relative to their source.
Pre-commit: the bundled hook id is supagents-build (see .pre-commit-hooks.yaml). It watches .agents/supagents/ and rewrites generated outputs in place.
# Generated by supagents from <rel-path>. Do not edit; edit the source instead. Edits to a marker-bearing file are overwritten on the next build. Hand-written siblings without the marker are preserved by clean.Code_Investigator.md, code_investigator.MD, or code investigator.md all fail validation. Lowercase letters, digits, -, _ only.README.md — full feature tour, pre-commit integration, configuration.src/supagents/config.py — target defaults, scope helpers, config schema.src/supagents/core.py — parsing, planning, idempotent writes (SOURCE_FILENAME_RE, marker constants).tests/fixtures/code_investigator.md — minimal source covering CLAUDE/GEMINI/COPILOT blocks plus an APPEND_BODY directive.Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub fmind/agent-supagents --plugin agent-supagents