The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from the project's codebase and documentation to generate an epistemic map.\n
Ingestion points: The skill automatically discovers and reads files from documentation directories (e.g., docs/, README.md) and source code directories (e.g., src/, lib/) as specified in SKILL.md.\n
Boundary markers: Although the skill uses a structured JSON schema for reporting findings, the instructions in SKILL.md do not mandate the use of delimiters or specific warnings to ignore instructions embedded within the ingested project content.\n
Capability inventory: Across SKILL.md and scripts/merge_epistemic_map.py, the skill has the capability to read local files, write scratch files and the final EPISTEMIC-MAP.md, and execute shell commands via a Python subprocess.\n
Sanitization: The merge script in scripts/merge_epistemic_map.py normalizes whitespace but does not implement sanitization or filtering to prevent malicious prompt instructions in the analyzed data from influencing the agent's output.\n- [COMMAND_EXECUTION]: The skill utilizes a bundled Python script to perform data merging and deduplication tasks.\n
Evidence: In SKILL.md, the agent is instructed to invoke python3 scripts/merge_epistemic_map.py with arguments pointing to absolute paths for findings and the output directory. This execution is part of the skill's primary purpose and is managed within the local environment.

epistemic-mapper