The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes an external Python script (lit_search.py) located in a sibling directory (~/.claude/skills/literature-search/). The execution command is constructed using parameters (lit_py, lit_skill_dir) defined in the loop.run.yaml configuration file.\n- [EXTERNAL_DOWNLOADS]: The skill instructions mention installing a missing sibling skill (literature-search). This implies fetching and executing external code that is not part of the skill's own package.\n- [PROMPT_INJECTION]: There is a risk of indirect prompt injection as the skill retrieves data from the web and academic sources (arXiv, Semantic Scholar) and processes it to extract claims and snippets without explicit sanitization or defensive boundary markers.\n
Ingestion points: Data enters the context via lit search, lit snippet, lit fulltext, and WebFetch operations as defined in SKILL.md.\n
Boundary markers: Absent; the instructions do not specify the use of protective delimiters when inserting retrieved snippets into the agent's context.\n
Capability inventory: The skill has the capability to execute subprocesses via python3 and write files to the local filesystem (matrix.json, survey.md, ledger.tsv).\n
Sanitization: Absent; the skill emphasizes extracting "verbatim" snippets, which may include malicious instructions from the source documents.

literature-survey