Agent skill · Audited from source
cold-start-interview
skill · anthropics/claude-for-legalApache-2.0 licenceAudited 13 Nov 2026 · source cb488a888d20
Is it safe? Yes — SAFE, grade B (89/100). We read this skill's source at commit cb488a888d20 and found no filesystem, network or shell access and no hard-coded secrets. What it does: Run the cold-start interview — learns your AI governance practice and writes `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md` from your AI policy, a reference impact assessment, and key vendor…
Cite
Cite this page
ModelsAtlas. “cold-start-interview (anthropics/claude-for-legal)”. Retrieved 13 Nov 2026. https://modelsatlas.com/skills/anthropics/cold-start-interview
Trust audit
Audit 0.4.1 · source read 13 Nov 2026
SAFE means nothing in the source contradicts what the skill says it does. Grade B is shared by 17,406 of 18,389 audited skills.
What the source does
- FilesystemNone observedNo reads or writes in the source
- NetworkNone observedNo outbound calls in the source
- ShellNone observedNo commands run in the source
- DependenciesPinnedNothing floats
- SecretsNone foundNo keys in source
Five audit layers
| Layer | What we check | Result | Findings |
|---|---|---|---|
L0 Provenance | Repo owner, licence, archive status, activity | PASS | 0 |
L1 Static source | Secrets, path traversal, supply chain, hidden code | NA | 0 |
L2 Instructions | Prompt injection, authority framing, hidden directives | PASS | 0 |
L3 Inventory | The files and declarations match the source | PASS | 0 |
L4 Sandbox | Run in isolation and watch behaviour | SKIPPED why | 0 |
Gates applied: no behavioural pass.
Findings (0)
No findings at any severity. We re-audit when the source changes.
Audit 0.4.1 · source sha cb488a888d20 · 13 Nov 2026Report an issue or request a re-scan
Provenance
Who published it, and what the verdict is pinned to
- Publisheranthropics GitHub organizationOwner of the repository this skill is read from
- Source folderanthropics/claude-for-legal/ai-governance-legal/skills/cold-start-interviewThe exact folder we audited
- LicenceApache-2.0As the repository states it
- Commit pinned for this verdictcb488a888d20A new audit runs when the source changes
- Repository statusNot archivedLast push 4 days ago
- Sandboxed runNot runSkills are read, not executed; layer L4 is skipped
What it tells the agent
Full source on GitHub ↗Its description: Run the cold-start interview — learns your AI governance practice and writes `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md` from your AI policy, a reference impact assessment, and key vendor…
---
name: cold-start-interview
description: >
Run the cold-start interview — learns your AI governance practice and writes
`~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md` from
your AI policy, a reference impact assessment, and key vendor AI agreements.
Use when the practice profile is missing or contains `[PLACEHOLDER]` markers,
or when user says "set up ai governance plugin", "onboard me", "configure ai
governance".
argument-hint: "[--redo | --check-integrations]"
---
# /cold-start-interview
1. Check `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md` — if populated and no `--redo`, confirm before overwriting.
2. Run the interview using the workflow below (includes Part 0 role + integration check).
3. Seed docs: AI/acceptable use policy (URL or file), a prior impact assessment, key vendor AI agreements, model inventory or allowlist/blocklist if they exist. Read all provided.
4. Extract: policy commitments and prohibitions, vendor positions (note gaps vs. stated), impact assessment structure, approved/prohibited tool lists.
5. Migration: if a populated CLAUDE.md (no `[PLACEHOLDER]` markers) exists at `~/.claude/plugins/cache/claude-for-legal/ai-governance-legal/*/CLAUDE.md` but not at the config path, copy it to the config path and tell the user what was migrated.
6. Write `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md` (create parent directories as needed). Show summary. Offer first task.
## Flags
- `--redo` — re-run the full interview and overwrite `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md`.
- `--check-integrations` — re-scan available MCP connectors and refresh the `## Available integrations` table in `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md` without re-running the full interview. Use after setting up a new connector (Slack, document storage, scheduled-tasks).
When probing: only report ✓ if an MCP tool call actually succeeded. Configured-but-untested connectors should be marked ⚪ with a one-line how-to for confirming. Never report ✓ based on `.mcp.json` declarations alone — that misleads users into thinking something is wired up when it isn't.
```
/ai-governance-legal:cold-start-interview
/ai-governance-legal:cold-start-interview --check-integrations
```
---
## Purpose
Learn how *this* AI governance team works — what role the company plays in the AI
supply chain, which regulations actually apply to them, what their red lines are for
AI use cases, and what good impact assessment looks like here. Write it into the plugin config
so every other skill reads from the same understanding.
AI governance postures vary enormously. A company that builds AI products for enterprise
customers has almost nothing in common with a company that deploys off-the-shelf AI
tools internally. The interview figures out which one this is before anything else —
because builder obligations and deployer obligations are nearly opposite exercises.
## Cold-start check
Read `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md`:
- **Does not exist** → start the interview.
- **Contains `<!-- SETUP PAUSED AT: -->`** → greet the user and offer to resume from that section.
- **Contains `[PLACEHOLDER]` markers but no pause comment** → the template was never completed; offer to start fresh or resume from wherever the placeholders begin.
- **Populated (no placeholders, no pause comment)** → already configured; skip unless `--redo`.
The template structure lives at `${CLAUDE_PLUGIN_ROOT}/CLAUDE.md` — use it as the section scaffold. Write the completed practice profile to the config path, creating parent directories as needed.
If a CLAUDE.md exists at the old cache path `~/.claude/plugins/cache/claude-for-legal/ai-governance-legal/*/CLAUDE.md` but not at the config path, copy it forward to the config path before proceeding.
## Check for the shared company profile
Look for `~/.claude/plugins/config/claude-for-legal/company-profile.md`.
- **If it exists:** Read it. Show a one-line confirmation: "You're [name], [practice setting], at [company], [industry], operating in [jurisdictions]. Right? (Or say 'update' to change the shared profile.)" If confirmed, skip the company questions — go straight to the plugin-specific ones.
- **If it doesn't exist:** You'll be the first plugin this user set up. After the orientation and fork, ask the company questions and write them to the shared profile (per the template at `references/company-profile-template.md` in the plugin root), then continue with the plugin-specific questions. Tell the user: "I've saved your company profile — the other legal plugins will read it and skip these questions."
The company questions that belong in the shared profile (and should NOT be re-asked if it exists): practice setting, company name, industry, what-you-sell, size, jurisdictions, regulators, risk appetite, escalation names. The plugin-specific questions (playbook positions, review framework, house style, supervision model, etc.) stay per-plugin.
## Install scope check
Before the orientation, if you notice the working directory is inside a project (not the user's home directory), flag it. Say once:
> **Heads up — it looks like this plugin may be project-scoped, which means I can only read files in [current directory]. If you'll want me to read documents from elsewhere (Downloads, Documents, Dropbox), install user-scoped instead — see QUICKSTART.md. You can continue with project scope, but you'll need to move files into this folder.**
Ask the user to confirm before proceeding: continue with project scope, or pause to reinstall user-scoped. If the working directory *is* the user's home directory, skip this check silently.
## Before the interview starts
Open with the fork-first preamble. Keep it to 3-4 short lines. Ask quick-or-full before anything else.
> **`ai-governance-legal` is for people who run AI governance: use-caExcerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.
Install
Pinned to the audited commit cb488a8
git clone https://github.com/anthropics/claude-for-legal git -C claude-for-legal checkout cb488a888d20
The author documents no install command, so this uses the repository directly. Our verdict covers commit cb488a888d20 only.
Works in which agents?
What its own text says, not a test
| Agent | Status | More |
|---|---|---|
| Claude Code | Mentioned in its text, not tested | Skills for Claude Code · About Claude Code |
| GitHub Copilot | Mentioned in its text, not tested | Skills for GitHub Copilot · About GitHub Copilot |
Audit history
Every audit, every source change
- ✓SAFE B 89 / 100
first audit · source
cb488a888d20
README badge
For maintainers · shows the latest audit
[](https://modelsatlas.com/skills/anthropics/cold-start-interview/)
The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.
Similar skills
mattpocock/skills
Implement a piece of work based on a spec or set of tickets.
mattpocock/skills
Compact the current conversation into a handoff document for another agent to pick up.
mattpocock/skills
Pursue a long-running goal in a single session by co-ordinating subagents.
Add this badge to your README
[](https://modelsatlas.com/skills/anthropics/cold-start-interview/)
<a href="https://modelsatlas.com/skills/anthropics/cold-start-interview/"><img src="https://modelsatlas.com/skills/anthropics/cold-start-interview/badge.svg" alt="ModelsAtlas: SAFE B"></a>
Questions about cold-start-interview
What does the cold-start-interview skill do?
Is cold-start-interview safe to install?
What can cold-start-interview access on my machine?
Which assistants does cold-start-interview work with?
How current is this page?
Sources
- github.com/anthropics/claude-for-legal/tree/main/ai-governance-l — the audited source at commit cb488a888d20 and every finding aboveObserved · 13 Nov 2026
Spotted something wrong? Report a correction · Methodology
Keep going in your AI
Opens with this page as the source. We add nothing else to the prompt.
- What does the cold-start-interview skill do?
- Is cold-start-interview safe to install?
- What can cold-start-interview access on my machine?