Agent Context Center

Static trust layer for AI-worked repos

Make AI context reviewable.

Agent Context Center turns AGENTS.md, CLAUDE.md, Cursor rules, MCP servers, hooks, commands, skills, docs, and TODOs into one offline dashboard that humans can open, diff, and commit.

View demo dashboard Install and run

acc --root . writes a source-linked dashboard.html into the owning provider folder.

Agent Context Center dashboard with overview, inventory, docs, TODOs, and cross-reference sections
A committed, source-linked dashboard reviewers can open without a server, account, or build step.

Use it where AI context can drift.

ACC is built for the review moments around coding agents, not for a marketing demo that only looks good once.

Review agent behavior

Regenerate the dashboard in a PR and inspect what changed across prompts, hooks, commands, MCP, and rules.

Onboard fast

Give a new maintainer or agent one source-linked map instead of a tour through hidden config folders.

Audit tools

See the declared MCP servers and hooks with redacted summaries, grouped by provider and source path.

Handoff cleanly

Commit a deterministic snapshot so the next person sees the repo's AI surface without trusting chat history.

Gate freshness

Use acc doctor --strict to flag stale dashboards and weak metadata in CI.

Explain publicly

Publish a static demo from GitHub Pages with no backend, telemetry, CDN, or runtime service.

Repo AI context files flow into dashboard.html and then into review, onboarding, audit, handoff, and CI use cases
Repo AI contextAGENTS.md, CLAUDE.md, Cursor rules, MCP, hooks, commands, docs, and TODOs.
dashboard.htmlOne static, redacted, source-linked map generated from the repo.
Developer workflowsReview PRs, onboard agents, audit tools, hand off repos, and gate CI drift.

How it works.

Markdown remains the source of truth. The dashboard is a generated read view for humans.

Scan the repo

Walks files deterministically while skipping local state, secrets, caches, and build artifacts.

Normalize providers

Claude, Codex, Cursor, and generic markdown become one shared inventory and docs schema.

Redact and validate

Secret-shaped values are redacted before rendering, then a tripwire checks the assembled output.

Commit the artifact

A self-contained dashboard opens offline and can be reviewed like any other generated file.

acc doctor --root . --strict

# deterministic checks
# stale dashboard
# weak metadata
# broken relative links
# redaction count
# open TODOs

Agents already know how to read the markdown. ACC gives humans the same operating picture, with source links and a file they can review.

Static HTML · deterministic output · stdlib Python 3.12+ · no telemetry

Security boundary

Review before publishing.

  • ACC redacts secret-shaped strings before rendering and re-scans the assembled output.
  • It is not a full entropy scanner, so unusual secrets still need human review.
  • The dashboard maps and summarizes repo context. It does not run agents, manage tools, or replace source markdown.