Agent context center

Skill omerakben/agent-context-center/skills/agent-context-center

Generate, refresh, or open this repo's Agent Context Center dashboard — a single self-contained offline HTML file that inventories the repo's AI configuration (agents, skills, hooks, commands, MCP servers, rules, docs, open TODOs). Use when the user asks to build, refresh, regenerate, or view that dashboard, and to re-stamp it after editing the repo's AI markdown when a dashboard.html already exists under a provider folder.From its SKILL.md

Install

npx -y skills add omerakben/agent-context-center --skill agent-context-center

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.9 KB, as published. Nobody here has run it

Agent Context Center

One committed HTML file mirrors a repo's AI markdown so a human can navigate it. Markdown stays the source of truth for the machine; the dashboard is the read view for the person. The generator is stdlib Python 3.12+, runs offline, and produces byte-stable output.

Generate or refresh

Run the /dashboard command. It selects a Python 3.12+ interpreter, runs the bundled generator against the user's project with no install and no network, and reports the written path, sourceDigest, and scanned-file count. The dashboard is written under the owning provider folder (.claude/, .codex/, or .cursor/), never the source tree, so it stays inert to web builds.

If more than one provider dashboard already exists, the generator stops and asks for an owner; re-run with --owner <dir>.

Keep it fresh (the default tier)

After you edit this repo's AI markdown (CLAUDE.md, AGENTS.md, an agent, skill, hook, command, MCP config, or rule) and a dashboard.html already exists under a provider folder, re-run /dashboard so the human view stays in sync. Refresh is manual by design: a static file:// page cannot tell it is stale. For automatic refresh, point the user at the opt-in templates in the plugin's templates/refresh/ (a git post-commit hook, a file-write hook snippet, and a CI drift check).

Enriching the dashboard — the only safe path

The generator writes every field, redacts secrets at extraction, and re-scans the assembled output with a tripwire. To add or improve a summary, edit the source markdown and re-run /dashboard so redaction and the tripwire re-fire over your text.

Never hand-edit dashboard.html or any generated file. A value written straight into the output bypasses redaction and the tripwire entirely.
When you write prose into source markdown, describe purpose and intent only. Never copy, quote, or paraphrase a literal value from a config file, an env block, a header, or a URL. The prose redactor is high-precision, not an entropy scanner, so a bare high-entropy token with no telltale prefix can slip through — do not rely on it to catch a value you pasted.

Invariants you must not undo

Deterministic output (re-stamping a clean repo produces no diff), redaction at extraction, and a textContent-only render are guarantees the generator enforces. Do not disable them, post-process the HTML, or inject raw markup. If a change would weaken any of them, stop and surface it instead.