ADR: Skill for Claude Code

Report a malicious skill

You prepare one GitHub issue on the ADR repository. The skill's zip is attached to the issue on GitHub after it is created.

1. Identify the skill

You need its folder hash (skill_sha256, 64 hex characters) and its zip.

• Take both from a Discovery snapshot: skill assets carry skill_sha256, install_path, provenance and, when the scan ran with --archive-skills, archive.path.
• No snapshot, or no archive? Scan with archiving on: adr-discovery --output-dir OUT --archive-skills (or python -m adr_discovery.cli ... from Discovery/), and read the skill from the newest OUT/snapshot-*.json.
• Check the zip is that skill. This must print the skill_sha256:

  python -c "import sys; from adr_discovery.archive import zip_content_sha256; print(zip_content_sha256(open(sys.argv[1], 'rb').read()))" <zip>
  

Ask the user in their own words why the skill is malicious, and for any public links that back it up.

If provenance shows a private or local marketplace, ask whether the skill belongs to their organisation. If it does, it should be reported there, not here.

2. Prepare the issue

Title: Malicious skill: <name> (<first 12 hex of skill_sha256>)

Body:

**Skill:** `<name>`
**Folder hash (`skill_sha256`):** `<skill_sha256>`
**Source:** <marketplace and plugin with version, and the marketplace repo, if public; otherwise "local or private">
**Seen at:** `~/<home-relative install path>`

**Why it is malicious**
<the user's reason>

**References**
- <https links, if any>

**Verify the attached zip**
`zip_content_sha256` of the zip must print the folder hash above:
`python -c "import sys; from adr_discovery.archive import zip_content_sha256; print(zip_content_sha256(open(sys.argv[1], 'rb').read()))" <skill_sha256>.zip`

**Proposed indicator**
```json
{"id": "skill-<name>-<first 12 hex>", "status": "active", "kind": "skill_sha256",
 "summary": "<one sentence: what the skill does>", "references": [...],
 "target": {"sha256": "<skill_sha256>"}}
```