project-config
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple internal framework scripts and shell commands using
nodeandnode -efor repository scanning, schema validation, and help generation. These commands are localized to the.claude/framework directory. It also performs administrative checks like verifying git ignore rules and potentially initiating environment repairs via WinGet on Windows systems. - [INDIRECT_PROMPT_INJECTION]: The skill manages
docs/project-config.jsonand~/.claude/.ck.json, which serve as the central control plane for the agent framework. Malicious configuration injected into these files could influence downstream agent behavior or tool routing. - Ingestion points: Reads
docs/project-config.json,~/.claude/.ck.json, and.claude/.ck.local.json(SKILL.md, local-workflow-routing.md). - Boundary markers: Uses structural schema validation via
.claude/hooks/lib/project-config-schema.cjsto enforce formatting, but does not employ specific LLM boundary delimiters (like XML tags or multi-shot examples) for the processed JSON content to isolate it from instructions. - Capability inventory: Performs filesystem writes to configuration files (SKILL.md Phase 3) and executes framework-specific CLI tools (SKILL.md Phase 0, help-mode.md).
- Sanitization: Implements schema validation and explicitly refuses to process paths targeting sensitive files like
.env,*.pem, or SSH keys (local-workflow-routing.md).
Audit Metadata