adr-discover
Pass
Audited by Gen Agent Trust Hub on Jul 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run local discovery commands including
find,grep, andlsto map the repository structure and identify design patterns in source code. These operations are limited to the local filesystem. - [DATA_EXFILTRATION]: The skill reads project configuration files (e.g.,
package.json,pyproject.toml,Cargo.toml) and source code to extract architectural metadata. While this exposes project structure to the agent, the information is used to generate local reports and no external exfiltration was identified. - [PROMPT_INJECTION]: The skill ingests untrusted content from the analyzed repository's files, creating a surface for indirect prompt injection. 1. Ingestion points: Project manifests and source code files (SKILL.md). 2. Boundary markers: None present to distinguish file content from instructions. 3. Capability inventory: Shell execution (
find,grep) and invocation of theadr-manageskill. 4. Sanitization: No explicit sanitization or escaping of ingested content. This risk is mitigated by the mandatory human-in-the-loop approval process for all generated candidates.
Audit Metadata