The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes GitHub issue descriptions to determine workflow readiness, creating an attack surface where malicious issue content could bypass validation logic.
Ingestion points: scripts/mark-ready.sh reads GitHub issue bodies via the gh CLI tool and passes them to a Python-based parser.
Boundary markers: None; the script treats the issue body as a single block of text for pattern matching.
Capability inventory: The skill can modify GitHub labels (gh issue edit) and execute local project scripts (plan-refresh.sh).
Sanitization: Content is parsed as JSON and evaluated against regular expressions, which can be bypassed by carefully crafted input satisfying the patterns while containing malicious intent.
[DYNAMIC_EXECUTION]: The skill uses dynamic execution techniques for its internal automation workflows.
scripts/mark-ready.sh executes a Python script provided through a shell heredoc.
The embedded Python script dynamically modifies its search path (sys.path) to import logic from a computed relative directory path (PARSER_DIR).
The shell script also includes logic to conditionally execute a file-based script (.agents/tools/plan-refresh.sh) if present in the project environment.

records