codeForerunner

Model-agnostic repository documentation tooling. Ships a prompt pack for codebase analysis and doc generation, a thin Python CLI, an MCP server, drift-detection rules that keep docs honest — and native slash-command skills for Claude Code, Codex, Gemini CLI, and other agent CLIs.

Two modes

Mode A — Agent skill (recommended, no API key required)

Install forerunner's prompt pack as skills into your agent CLI. Each documentation task becomes a slash command (/forerunner-readme, /forerunner-check, etc.) available inside Claude Code, Codex, Gemini CLI, and other agents. Authentication is handled by your existing agent subscription — no separate API key needed.

# From a cloned repo
./install.sh

# One-liner (auto-detects Claude Code, Codex, Gemini CLI)
curl -fsSL https://raw.githubusercontent.com/derek-palmer/codeforerunner/main/install.sh | bash

# Windows
irm https://raw.githubusercontent.com/derek-palmer/codeforerunner/main/install.ps1 | iex

# Via forerunner CLI (after pip install)
forerunner install --all claude
forerunner install --all codex

Then in your agent:

/forerunner-scan        ← scan the repo first
/forerunner-readme      ← generate README
/forerunner-check       ← detect doc drift

Mode B — Direct API (needs API key or Ollama)

Install the Python CLI and call your provider directly. Works without any agent CLI installed.

pipx install codeforerunner   # recommended
pip install codeforerunner    # alternative

Configure a provider (or start Ollama for keyless local generation):

export ANTHROPIC_API_KEY=sk-...
forerunner generate readme --stream

If no API key and no --provider flag, forerunner auto-detects Ollama at localhost:11434 and falls back to local mode.

Slash commands

Command	Task	Purpose
/forerunner-scan	scan	Collect repo evidence (run first)
/forerunner-readme	readme	Generate or refresh README.md
/forerunner-api-docs	api-docs	Generate API reference docs
/forerunner-diagrams	diagrams	Generate Mermaid architecture diagrams
/forerunner-flows	flows	Document system flows
/forerunner-stack-docs	stack-docs	Stack-specific developer docs
/forerunner-version-audit	version-audit	Audit pinned versions vs EOL
/forerunner-check	check	Check docs for staleness
/forerunner-review	review	Doc-impact summary for PR review
/forerunner-audit	audit	Security and dependency audit
/forerunner-changelog	changelog	Generate changelog from git log
/forerunner-init	init-agent-onboarding	Bootstrap or refresh AGENTS.md

Slash command availability depends on the agent CLI. Claude Code, Codex, and Gemini CLI support all 12 commands after install.

Skill install options

Flag	Effect
./install.sh	Auto-detect all agents, install all skills
./install.sh --only claude	Claude Code only
./install.sh --only codex	Codex only
./install.sh --only gemini	Gemini CLI only
./install.sh --dry-run	Preview, write nothing
./install.sh --list	Show detected agents + skill list
./install.sh --uninstall	Remove all installed skills

CLI

pip install codeforerunner

Command	Purpose
forerunner init	Resolve agent-onboarding bundle to stdout.
forerunner scan	Resolve scan bundle to stdout.
forerunner doc <task>	Resolve base + partials + task bundle to stdout.
forerunner check	Run drift-detection rules; no-op without forerunner.config.yaml.
forerunner generate <task>	Call configured provider directly. Add --stream for token-by-token output. Falls back to Ollama automatically when no API key is configured.
forerunner doctor	Health report: skill parity, config, provider key, local-mode status. Add --fix to write a starter config.
forerunner mcp-server	Serve prompt bundles as MCP tools over stdio (JSON-RPC 2.0).
forerunner install <agent>	Install canonical skill into agent-specific directory. Add --all for all per-task skills.

Prompt pack

Prompts are bundled inside the package at src/codeforerunner/prompts/.

prompts/
├── system/base.md
├── partials/
│   ├── context-format.md
│   ├── output-rules.md
│   └── stack-hints.md
└── tasks/
    ├── scan.md          api-docs.md    audit.md
    ├── readme.md        diagrams.md    changelog.md
    ├── check.md         flows.md       version-audit.md
    ├── review.md        stack-docs.md
    └── init-agent-onboarding.md

Quick start (agent skill mode)

# Install skills into Claude Code
curl -fsSL https://raw.githubusercontent.com/derek-palmer/codeforerunner/main/install.sh | bash

# In Claude Code:
# /forerunner-scan    → scans your repo
# /forerunner-readme  → generates README.md
# /forerunner-check   → checks for doc drift

Quick start (direct API mode)

# 1. Install and configure
pip install codeforerunner
export ANTHROPIC_API_KEY=sk-...

# 2. Run a task
forerunner generate readme --stream

# 3. Enable drift detection
forerunner doctor --fix        # writes forerunner.config.yaml
forerunner check               # run any time or as pre-commit hook

GitHub Action

- uses: derek-palmer/codeforerunner@v0.3.2

No-op when forerunner.config.yaml is absent.

Configuration

Copy forerunner.config.yaml.example to forerunner.config.yaml to opt in to drift rules. Generate a starter config with:

forerunner doctor --fix

Config fields

provider: anthropic          # anthropic | openai | google | ollama
model: claude-opus-4-7
api_key_env:
  anthropic: ANTHROPIC_API_KEY

tasks:
  check:
    enabled_rules:
      - R1-no-cli
      - R2-no-pre-commit
      - R3-no-ci
      - R4-no-installer
      - R5-no-python-package
      - R7-no-mcp
      - R8-no-marketplace
      - RI1-missing-cli
      - RI5-missing-python-package
      - RI7-missing-mcp
      - RV1-version-drift
    ignore_paths:
      - docs/legacy/**/*.md

Drift rules

Rule	Fires when
R1-no-cli	Doc denies having a CLI, but cli.py is present
R2-no-pre-commit	Doc denies having pre-commit hooks, but .pre-commit-hooks.yaml present
R3-no-ci	Doc denies having CI, but .github/workflows/*.yml present
R4-no-installer	Doc denies having an installer, but installer.py present
R5-no-python-package	Doc denies having a Python package, but pyproject.toml present
R6-no-docker	Doc denies having Docker, but Dockerfile/compose.yml present
R7-no-mcp	Doc denies having an MCP server, but mcp_server.py present
R8-no-marketplace	Doc denies having a marketplace, but marketplace.json present
RI1-missing-cli	Doc references forerunner subcommands but cli.py absent
RI5-missing-python-package	Doc shows pip install codeforerunner but pyproject.toml absent
RI7-missing-mcp	Doc references forerunner mcp-server but mcp_server.py absent
RV1-version-drift	Doc pins codeforerunner==X.Y.Z differing from current version

MCP Server

forerunner mcp-server speaks JSON-RPC 2.0 over stdio and exposes one tool per prompts/tasks/*.md. A scan-first gate enforces SPEC V2: any tool except scan or init-agent-onboarding returns an error until scan has been called in the same session.

See examples/mcp/ for Claude Desktop and mcp-cli wiring examples.

Providers

forerunner generate supports four providers. When no provider is explicitly configured and no API key is found, forerunner probes localhost:11434 and falls back to Ollama automatically.

Provider	Env var	Default model
anthropic	ANTHROPIC_API_KEY	claude-opus-4-7
openai	OPENAI_API_KEY	gpt-4o
google	GOOGLE_API_KEY	gemini-2.5-pro
ollama	OLLAMA_HOST (optional)	llama3

Docs and spec

• SPEC.md — canonical phase/task tracker
• docs/getting-started.md — manual prompt use
• docs/prompt-guide.md — how system, partial, and task prompts compose
• docs/editor-agent-setup.md — adapting prompts to local agents
• docs/roadmap.md — human-readable roadmap
• docs/agent-distribution-design.md — packaging and installer design