CLI Commands

Command	Purpose
agentdocs try <url-or-path> --goal <goal>	Build, audit, and find context in one command
agentdocs context <goal>	Produce a compact agent context bundle from built artifacts
agentdocs handoff <goal>	Produce an agent-native task handoff with freshness and MCP guidance
agentdocs setup-agent	Print MCP setup snippets for common coding-agent clients
agentdocs status	Check whether generated artifacts are fresh
agentdocs rebuild --changed	Recollect stale configured sources and rebuild artifacts
agentdocs watch	Poll status and rebuild when configured sources become stale
agentdocs verify-context --task <goal>	Check task context for staleness, conflicts, weak evidence, and mismatches
agentdocs init	Create starter configuration
agentdocs ingest <path>	Collect local Markdown, MDX, Sphinx/reST, and AsciiDoc
agentdocs crawl <url>	Collect same-origin website documentation
agentdocs build	Generate artifacts and search index
agentdocs doctor	Audit agent readiness
agentdocs search <query>	Search built artifacts offline
agentdocs inspect entities	Inspect extracted entities
agentdocs inspect links	Inspect extracted relationships
agentdocs inspect task-pack <id>	Explain why a task pack was generated
agentdocs export --format <format> --to <path>	Export generated artifacts
agentdocs serve-mcp	Start the read-only MCP server

Global options include --config, --out, --cwd, --json, --quiet, and --verbose.

agentdocs --help
agentdocs try ./docs --goal "implement authentication"
agentdocs handoff "implement authentication"
agentdocs setup-agent --client codex
agentdocs status
agentdocs build --help
agentdocs build --check
agentdocs build --clean
agentdocs export --format llms --to ./public --force

Additional inspect targets are planned.

build --clean safely removes the configured AgentDocs output directory before collecting and rebuilding. It refuses to clean the project root, filesystem roots, or paths outside the working directory.

build --check is a non-mutating CI drift gate. It reads state/build-state.json, compares configured source fingerprints and generated artifact hashes, and fails when context is stale, missing, or unknown. It does not crawl, collect, prune, rebuild, or write files. --json emits the same status report shape as agentdocs status --json.

export --format static copies the complete built output directory. export --format llms copies the publishable agent-facing subset: llms.txt, generated AGENTS.md, agent-brief.md, manifest.json, agent-map.json, chunks.jsonl, task-packs/, and readiness reports when present. Export refuses destinations inside the active output directory unless they are moved elsewhere first, and --force is required to replace a non-empty destination.

Task-pack inspection reads the validated agent-map.json and reports the pack's confidence, required pages, source evidence, steps, and related entities:

agentdocs inspect task-pack quickstart
agentdocs --json inspect task-pack authentication

try accepts --include, --exclude, --max-pages, and --sitemap for scoped website trials. Direct crawl and try infer a guide scope unless explicit include patterns are supplied.

ingest --strict disables tolerant MDX fallback. search --facet key=value is repeatable and applies hard context filters:

agentdocs ingest ./docs --strict
agentdocs search "migration" --facet version=v5
agentdocs search "route handler" --facet framework=nextjs --facet router=app

Search JSON includes evidence-linked facets and deterministic context_conflict warnings when unfiltered top results mix exclusive context.

handoff is the recommended multi-session command. context remains supported for the earlier compact bundle shape. status, rebuild --changed, and watch use deterministic source fingerprints for local/repo sources and a website TTL for crawled sources.