The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/write_outputs.py dynamically loads a shared module write_explore_bundle.py from a relative path (../../../../shared/scripts/). While this appears to be a standard pattern for shared vendor utilities in this environment, it involves dynamic execution of code located outside the skill's own directory.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it processes content from external repositories and user-provided JSON configurations to plan and summarize code changes.
Ingestion points: scripts/plan_code_changes.py reads data from paths provided via --variant-spec-json, --idea-card-json, and --analysis-json, and it performs a recursive scan of the repository provided via the --repo argument.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands were found in the processing logic.
Capability inventory: The skill has capabilities to read files and execute scripts.
Sanitization: Input data is parsed as JSON but the resulting content is used to build instructions and plans without explicit sanitization to prevent the interpretation of embedded malicious prompts.

explore-code