harness-step1-create-agents-md
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes standard Unix commands such as
find,cat,ls, andheadto gather information about the project's directory structure, dependencies, and existing documentation. These operations are conducted in a read-only manner and are confined to the local project environment. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from project files, including
README.mdand package manifests (package.json,pyproject.toml, etc.), to generate summaries. This creates a potential surface for indirect prompt injection if these files contain malicious instructions meant to influence the agent's output during documentation generation. - Ingestion points: The skill reads file metadata and content from
package.json,pyproject.toml,go.mod,Cargo.toml, and the first 80 lines ofREADME.md. - Boundary markers: There are no specific instructions to use delimiters or ignore-embedded-instructions warnings when processing external file content.
- Capability inventory: The skill is designed to write multiple markdown files to the
docs/directory and create anAGENTS.mdfile in the root. - Sanitization: The agent is directed to extract specific business and technical details rather than performing a raw copy of the ingested text.
Audit Metadata