The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses shell scripts (ultima.sh and tickets.sh) that contain embedded Python code executed via heredocs. This approach is used for complex data processing and API interactions while remaining transparent and using only the Python standard library.
[COMMAND_EXECUTION]: The skill invokes standard system utilities such as git, grep, and find, along with the GitHub gh CLI. These tools are used for profiling the repository and searching for code patterns relevant to the audit, which is the core functionality of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill audits external codebase files, creating a theoretical surface for indirect prompt injection.
Ingestion points: Frontend source code files (.tsx, .jsx, .vue, .svelte, .astro, .css, etc.) are read during Stage 4 analysis.
Boundary markers: The references/lens-template.md defines a strict prompt template and "Rules of engagement" that instruct subagents to ignore irrelevant data and follow a specific JSON output contract.
Capability inventory: The skill possesses capabilities for reading files, executing shell commands for analysis, and performing authenticated network writes to issue trackers.
Sanitization: All audit findings are validated against a strict JSON schema (candidates-schema.json), and the final HTML report is generated by a script that escapes all fields and explicitly blocks the inclusion of <script> tags.
[SAFE]: The skill implements robust security controls for its temporary work environment, creating directories with mkdir -m 700 and performing ownership and symlink checks to prevent local privilege escalation or file manipulation attacks. Credentials for issue trackers are handled through standard environment variables or CLI tools without being hardcoded.

ultima