Skip to content
Back to skills

Shaman

ASecurity

Have the AI run a read-only orchraft setup preflight. Use when the user says shaman, asks whether orchraft is ready, or wants a preflight before another workflow.

  • 2 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 3, 2026
ai-agentsrailsgit

Security analysis

A100/100

Scanned October 3, 2026

npx -y skills add karloows/orchraft --skill shaman --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Shaman?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Shaman
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/karloows-shaman/badge)](https://www.skillsdirectory.com/skills/karloows-shaman)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: shaman
description: Have the AI run a read-only orchraft setup preflight. Use when the user says shaman, asks whether orchraft is ready, or wants a preflight before another workflow.
---

# Shaman Workflow

Use this skill before `ship`, `roast`, or `land` when setup is uncertain. It
reads the signs of the current repository; it never repairs anything.

## At A Glance

1. Resolve the current repository root and the orchraft plugin root.
2. Run the checks below without changing files, git state, GitHub, or Trello.
3. Report each check as `PASS`, `WARN`, or `FAIL`, then give the smallest next
   action and an explicit `READY` or `NOT READY` result.

## Checks

### Repository and skill loading

- Confirm `git rev-parse --show-toplevel` succeeds and report the repository
  root and current branch.
- Confirm the loaded orchraft package contains `skills/` and the skill files
  named by the requested workflow. If this skill is running, its own loading
  is a pass; inspect the package for the other requested skills rather than
  claiming every skill is available.
- Confirm the repository has an `origin` remote when a GitHub workflow is
  requested. Do not fetch, push, or alter remotes.

### Policy fallback

- For `approval-policy.md`, `config-policy.md`, and
  `verification-policy.md`, report whether the target repository provides a
  local copy or orchraft will use the bundled fallback.
- If a local policy exists, read it as the target repository's authority. A
  missing local policy is a warning only; fallback is supported.

### GitHub access

- If `gh` exists, run `gh auth status` and report its result without exposing
  tokens. A missing or unauthenticated `gh` is a warning when the configured
  GitHub connector can provide read access, otherwise a failure for workflows
  that need GitHub state.
- Confirm the origin URL is a GitHub repository when GitHub work is requested.
  Do not treat a successful local git command as proof of GitHub
  authentication.

### Required tools

- `git` is required for every workflow.
- `jq` is required for the bundled hooks; missing `jq` is a warning that
  ambient nudges will be silent.
- `gh` is required only when no authenticated GitHub connector is available.
- Report missing optional tools plainly instead of inventing a workaround.

### Config parsing

- Check `.orchraft.jsonc` first, then `.orchraft.json`, and ignore both when
  neither exists because config is optional.
- Parse the selected file as JSON. For `.jsonc`, use the same quote-aware
  comment stripping convention as `hooks/watchtower-nudge.sh`; never strip
  `//` inside a quoted URL.
- A malformed file is a `WARN` that names the file and parse error, then
  explain that workflows will continue with documented defaults. Do not edit
  or rewrite the file.
- For a valid file, check only the documented types that affect the requested
  workflow. Unknown keys are not failures.

### Hook support

- If `hooks/hooks.json` is present, validate it as JSON, confirm every command
  it references exists, and confirm referenced scripts are executable.
- Confirm `jq` is available for the bundled hook scripts.
- Report unsupported host hook behavior as a warning with the host-specific
  limitation. Never install, enable, or modify a hook.

## Output

Use this compact format:

```text
Repository: <root> (<branch>)
<status>  Skills: <result>
<status>  Policies: <local/fallback result>
<status>  GitHub: <authenticated / connector / unavailable>
<status>  Tools: <result>
<status>  Config: <absent / valid path>
<status>  Hooks: <supported / not installed>

Next action: <smallest action, or none>
Result: READY | NOT READY
```

Use `FAIL` for a broken prerequisite or malformed config; any `FAIL` in a
requested workflow means `NOT READY`. Use `WARN` for a supported fallback or
unavailable optional integration; a warning permits `READY` only when the
requested outcome still works through that fallback or the warning concerns
an unavailable optional integration. Use `PASS` only for a check actually
performed in this turn. Never claim a check passed from memory.

## Guardrails

- Read-only: never create or update a branch, commit, push, fetch, open or
  update a PR, change an issue or release, mutate Trello, or edit files.
- Do not run commands that write caches, generated files, credentials, or
  dependency locks.
- Do not print secrets, tokens, full credential paths, or config values that
  may contain secrets.
- Stop after the report. The shaman may recommend `warplan`, `forgehand`,
  `ship`, `roast`, or `land`, but it never invokes the next workflow.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.