# Adr

> |

- **Type:** Skill
- **Install:** `agentstack add skill-beevibe-ai-beevibe-cto-claude-code-skill`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [beevibe-ai](https://agentstack.voostack.com/s/beevibe-ai)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** Apache-2.0
- **Upstream author:** [beevibe-ai](https://github.com/beevibe-ai)
- **Source:** https://github.com/beevibe-ai/beevibe-cto/tree/main/examples/claude-code-skill
- **Website:** https://beevibe.ai/cto

## Install

```sh
agentstack add skill-beevibe-ai-beevibe-cto-claude-code-skill
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# /adr — Architecture Deep Research

When the user invokes `/adr` (or asks any of the trigger questions in the description above), do the following.

## Step 1. Confirm the decision name

Ask the user one question, in chat:

> What's the architecture decision you're making? (e.g. "event bus topology", "retrieval architecture", "auth provider")

Capture their answer as ``.

If the user already named the decision when they invoked the skill, skip the question.

## Step 2. Run discover-first deep-research via the MCP server

Call the `adr_deep_research` MCP tool with these arguments:

```json
{
  "discover_first": true,
  "repo_path": ".",
  "domain": "",
  "decision": "",
  "out_dir": ".adr-runs/"
}
```

This will:

1. Scan the user's repo and draft a PRD (no network calls).
2. Run the full ADR pipeline against the draft (research, knowledge map, comparison matrix, synthesis, citation audit, evaluation pack).
3. Return the parsed `execution-handoff.json` so you can summarize the decision.

A run typically takes 3–6 minutes. Tell the user roughly how long it'll take before calling the tool so the wait doesn't feel like a hang.

## Step 3. Summarize the result

The tool response includes:

- `handoff.selected_topology` — the chosen architecture family
- `handoff.required_invariants` — non-negotiable constraints
- `handoff.forbidden_topologies` — what NOT to do
- `handoff.critique_summary.recommend_human_review` — if true, the kernel is telling you the decision is borderline
- `handoff.comparison_matrix_summary` — candidate count, empty cells
- `handoff.citation_audit_summary` — how many citations verified

Show the user a 3–5 line summary:

```
Selected: 
Required: 
Avoid:    

```

Then offer to:
- Open `ADR.md` for the full human-readable decision record
- Walk through the comparison matrix
- Implement using `execution-handoff.json` as the contract

## Step 4. (optional) Implement under the handoff

If the user says "go ahead and implement," read `/execution-handoff.json` and treat it as a hard contract:

- Honor `required_invariants` in the code you write
- Never reach for anything in `forbidden_topologies`
- Run against `domain-evaluation-pack.json` test cases before declaring done

## Failure modes

- **No LLM provider configured**: the tool will return an isError result. Tell the user to `export ADR_OPENAI_API_KEY=...` (or `OPENAI_API_KEY`) and re-invoke.
- **No live search provider configured**: same as above, but for `BRAVE_SEARCH_API_KEY` / `TAVILY_API_KEY` / `SERPER_API_KEY` / `SEARXNG_URL`, OR the OpenAI key fallback for hosted `web_search`.
- **`recommend_human_review: true`**: do NOT proceed to implementation. Show the user the borderline and ask whether to accept the decision, override it, or run a superseding ADR with a tighter brief.

## Notes for Claude

- The MCP tool name is `adr_deep_research`. Call it through the MCP host's tool-call mechanism — do not try to spawn a subprocess.
- The skill assumes the `adr` MCP server is registered in the user's Claude Code config. If it isn't, point them at `examples/claude-code-skill/.mcp.json` in the beevibe-cto repo.
- For quick scans without the full deep-research run, use `adr_discover` instead. It returns only the draft PRD and skips the live-research loop.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [beevibe-ai](https://github.com/beevibe-ai)
- **Source:** [beevibe-ai/beevibe-cto](https://github.com/beevibe-ai/beevibe-cto)
- **License:** Apache-2.0
- **Homepage:** https://beevibe.ai/cto

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** yes
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-beevibe-ai-beevibe-cto-claude-code-skill
- Seller: https://agentstack.voostack.com/s/beevibe-ai
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.