Outfitter community agents

Synced from community-profiles/README.md. The repository is the source of truth.

Community-contributed Dotagents catalog for Outfitter. Agents and skills here are reviewed for structure but are not curated like the default catalog — read an agent and its selected resources before you run it.

Agents

actions-agent - conventional headless identity for GitHub Actions automation.
founder - owns the mission, priorities, constraints, and final decisions.
planner - maintains plans, summarizes project status, delegates work, and writes daily reports. It does not implement changes.
engineer - owns implementation and verification.
product-marketer - owns outbound communication; turns merged change into user stories and release notes.
researcher - produces sourced research, maintains durable knowledge, authors personas, and reviews artifacts from a persona. It does not select a runtime environment.
explorer - maps repositories or systems in a read-only environment.
environment - abstract repository location and authentication baseline for inherited agents.
environment.repos - abstract repository checkout and worktree layout under ~/repos: one clean clone on the default branch, all work in typed worktrees.
environment.agent-operator-pod - abstract runtime context for resident agents in Kubernetes pods.
environment.actions-runner - abstract runtime context for ephemeral CI agents.
environment.container-readonly - abstract tool boundary for read-only agents.
environment.repo-auth - abstract, replaceable repository transport and forge authentication convention.
git-forge-delegator - creates and reviews delegated forge work. Template profile: compose it (inherits: [environment, git-forge-delegator]), do not run it directly — it no longer carries the repository and auth environment itself.
platform - platform engineering setup for infrastructure, CI/CD, deployment, reliability, browser-debugging evidence, and developer tooling.
media-editor - video post-production setup for transcript-driven editing with whisper.cpp and ffmpeg. See docs/media-editor.md.
persona-reviewer - one shared review profile whose identity is supplied by persona Markdown appended at launch.
grafana-agent - platform agent for Grafana on Kubernetes: provisions the Grafana MCP securely and declaratively, and investigates firing alerts with read-only, comment-only diagnosis. Ships agent-local skills (grafana-mcp-setup, grafana-alert-investigate, alert-issue-triage) and a per-agent mcp.json merged into the composition. See docs/grafana-agent.md.

Skills

browser-mcp - use Playwright or Chrome DevTools MCP for browser automation, UI debugging, screenshots, console/network inspection, and isolated browser sessions.
persona-authoring - create a self-contained persona document whose details live in portable Markdown.
persona-review - run one isolated shared reviewer with a canonical persona file and save a sourced report in the adopted voice.
media-editor - transcript-driven video editing: toolchain setup, whisper.cpp transcription, and ffmpeg cut/speed/export, with per-step references.
pyramid-principle - structure ideas, documents, and communications top-down (conclusion first) for clarity.
code-review - review a pull request diff against its issue’s acceptance criteria; approve, request changes, or merge when green.
prose-review - review prose artifacts for thesis, structure, and register before they publish or merge.
issue-triage - classify and comment on new GitHub issues.
mermaid - generate Mermaid diagrams across 20+ diagram types, routing to a per-type syntax reference. Vendored from WH-2099/mermaid-skill (MIT).
project-daily-report - collect project and telemetry evidence, write a linked daily Markdown report, and publish one idempotent edition.
slidev - scaffold a minimal Slidev deck: package.json, slides.md, dev/build/PDF-export scripts. Setup only.
storyboard - sequence a story into beats, each pairing a claim, a prose line, and an image prompt, written as storyboard.md before deck or one-pager work.
replicad - code-CAD with replicad v0.19 on the OpenCASCADE B-rep WASM kernel: parts as functions, subassemblies, STL/STEP export, and a minimal side-by-side HTML viewer.

Prompts

Shared prompt fragments in prompts/, appended by agents via append_system_prompt: [{file: ...}]:

prose.simplified-technical-english - the succinct ASD-STE100 base register (engineer, planner).
prose.rfc2119-requirements - requirements and acceptance criteria with RFC 2119 keywords (planner, founder, git-forge-delegator).
practice.draft-pr-lifecycle - author changes through a draft PR: iterate and verify CI drafted, mark ready when green, enable auto-merge via the merge queue (engineer, product-marketer).
practice.adversarial-review - review to find the failure; anchor findings as inline comments on real diff line numbers (engineer, product-marketer).

See Persona documents and reviews for the setup and runtime boundary.

See Environment baseline for the composable repository convention and the role-family boundaries.

See Scaling by composition for how an organization grows this catalog’s roles: skills first, then resident fission with forge-mediated coordination.

Using this catalog

Add the standalone catalog as a source and sync:

sources:
  - github: ai-outfitter/community-profiles
    ref: <tag-or-commit>

outfitter sync
outfitter run <agent-id>

Pin ref to a tag or commit — an unpinned source runs whatever the catalog publishes next.

Workflows

workflows/<id>/workflow.yaml contains the ten canonical, declarative workflow graphs. Outfitter resolves and validates their complete agent, skill, prompt, MCP, nested-workflow, and pinned-artifact closure; it never executes workflow nodes. Publication is blocked unless outfitter validate --strict succeeds.

To validate against a local Outfitter build:

scripts/validate-workflows.sh /path/to/outfitter/code/cli/dist/cli.js

Contributing an agent or skill

Add agents/<id>/agent.md with a matching name, a precise description, and the smallest useful loadout.
Keep durable policy in the agent and reusable procedures in skills/<id>/SKILL.md; prefer adding a skill over a near-duplicate agent.
A template profile (a capability bundle meant for composition, like git-forge-delegator) MUST NOT inherit other profiles; the runnable profile composes the full chain — see Scaling by composition.
Select resources by slug from agent frontmatter. Keep Pi-only extensions explicit and pinned when reproducibility matters.
Run outfitter validate --strict, then open a pull request that names the intended harnesses (Pi or Claude Code).

Prefer opening an issue first: newly opened issues are triaged automatically by the conventional actions-agent and its selected issue-triage skill (running on this repo via ai-outfitter/actions, see .github/workflows/issue-triage.yml). The agent labels the issue feat (new agent, skill, or prompting change) or fix, and comments with a suggested plan and example sketches following Outfitter best practices. Issues it cannot classify confidently get no label — just a comment asking a maintainer to take a look.

Layout

agents/         one directory per agent identity and loadout
skills/         reusable Agent Skills packages
prompts/        shared prompt fragments appended by agents
workflows/      declarative workflow packages resolved by Outfitter
models.json     model/provider configuration used by the CI agent
settings.yml    Outfitter defaults for this standalone payload