The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The render-index.py script processes metadata from record.md files which may be provided by untrusted sources.
Ingestion points: The script traverses the records/ directory and reads YAML-like frontmatter from every record.md file found.
Boundary markers: The generated INDEX.md output is structured as a Markdown table and includes a clear generator notice: .
Capability inventory: The skill provides file-writing capabilities via safe-write.py and reads repository metadata.
Sanitization: The script implements a literal() function that performs comprehensive sanitization. It applies html.escape() and backslash-escapes Markdown syntax characters (\, `, *, _, [, ], <, >, |) before outputting content, effectively preventing the injection of malicious formatting or instructions into the agent's context.
[COMMAND_EXECUTION]: Static analysis identified the use of subprocess.run within the project.
Evidence: The calls occur in scripts/test_render_index.py and scripts/test_safe_write.py.
Context: These calls are limited to the test suite and are used to invoke the skill's own internal scripts using the current Python interpreter (sys.executable). They do not ingest unvalidated external input and are used for functional validation only.

akosile