The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (wiki/ and raw/ directories) to synthesize answers and generate new wiki content, creating a surface for adversarial instructions in the processed data to influence agent behavior.
Ingestion points: Data is read from files within the wiki/ and raw/ paths (SKILL.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands within ingested data are defined.
Capability inventory: The skill can execute local Python scripts and perform file write operations to the wiki/ directory (SKILL.md).
Sanitization: There are no documented sanitization or filtering steps for the content ingested from external source files.
[COMMAND_EXECUTION]: The skill relies on executing multiple local Python scripts (mem-rank.py, query-log.py, wiki-graph.py, wiki-sync.py) and the ripgrep (rg) utility to perform search, ranking, and citation tasks.
[DATA_EXFILTRATION]: The skill contains a telemetry step (query-log.py --record) that logs user questions, accessed pages, and token usage metadata. While described as part of a retrieval evaluation framework, such telemetry mechanisms track interaction data which is a privacy consideration.

query