ai-context-refresh
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill performs local command execution to update project documentation and synchronize Codex mirrors. It invokes Node.js scripts such as
generate-claude-md.cjsand the standalone runnerrun-codex-sync.mjs. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted project files and interpolates it into the system-level instruction file
CLAUDE.md. - Ingestion points: Reads project facts and descriptions from
docs/project-config.jsonvia theloadConfigfunction ingenerate-claude-md.cjs. - Boundary markers: Generated content is encapsulated within explicit markdown comments, such as
<!-- SECTION:key -->and<!-- /SECTION:key -->, to prevent accidental interference with hand-authored text. - Capability inventory: The skill has the ability to write to the file system (
fs.writeFileSync) and execute local shell commands via Node.js. - Sanitization: The builder implementation includes markdown escaping for structural characters (using
tableCell,codeCell, andliteralfunctions) and an explicit redaction protocol (renderCredentialReference) that filters out sensitive data like API keys, private keys, and tokens using a set of comprehensive regex patterns.
Audit Metadata