Human ↔ agent documentation bridge

One repository.
Two audiences.
Zero duplicated truth.

Doc Bridge turns ownership into deterministic handoffs, MCP tools, freshness gates, and reviewable memory promotions — before any model burns tokens on the wrong files.

install
pnpm add -D @agentskit/doc-bridge
# or: npm i -D @agentskit/doc-bridge
Run the 60-second proof Memory → docsMCPllms.txt
Layer 0
No API key
MCP
Same handoffs
Gates
Fail-closed CI
deterministic handoff
$ npx ak-docs demo --text
# 60s: handoff · gate red→green · MCP snippet

$ ak-docs index
✓ 29 knowledge sources indexed

$ ak-docs query package doc-bridge --agent
{
  "startHere": "docs/POSITIONING.md",
  "editRoots": ["src"],
  "checks": ["pnpm test", "pnpm typecheck"]
}

backend calls: 0

Live corpus · 29 indexed public sources · copy any command

Why it wins

Not another wiki. A bridge that agents can execute.

Deterministic first

Known questions never burn tokens. Provenance stays in the repository.

Fail-closed gates

CI refuses drift. Agents inherit the same truth as humans.

MCP-native

Cursor, Claude, Codex — same handoffs as the CLI, not a second corpus.

Memory → docs

Digest agent notes, classify, promote as reviewable drafts — never silent merge.

The proof

Useful before AI enters the loop.

Known questions stay local and fast. Ambiguity surfaces with provenance. Only an unresolved miss earns a backend call.

Full CLI map
  1. 01

    Index

    command
    ak-docs index

    Build a deterministic knowledge map from source docs — offline.

  2. 02

    Resolve

    command
    ak-docs query package auth --agent

    Exact handoff JSON: startHere, editRoots, checks.

  3. 03

    Gate

    command
    ak-docs gate run

    Fail stale context before incomplete prompts hit a model.

Human ↔ agent loop

A feedback system — not another documentation silo.

Human docs

Explain intent and operation.

Ownership map

Source, checks, edit roots.

Agent handoff

Compact deterministic context.

Memory promote

Draft PR from durable learnings.

Continue with context

One ecosystem. Six useful next steps.

Choose the next constraint you actually have. Every product remains optional and links back into the same workflow.

01AgentsKitStart with the JavaScript foundation library.Foundation library for agents and applications. ↗02AgentsKit RegistryBrowse production-ready agent source.Start from reusable agents. ↗03AgentsKit ChatBuilding chat? Use the cross-framework application layer.Deliver conversational interfaces. ↗04Agents PlaybookKeep agent-built code safe and shippable.Apply repeatable engineering standards. ↗05AgentsKit Code ReviewRun a focused, low-noise model review.Verify changes before merge. ↗06AgentsKit OSNeed an enterprise solution? Add orchestration, policy, and audit.Operate and govern in production. ↗