The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from markdown frontmatter to perform taxonomy analysis, which could allow a malicious file to influence the agent's report or migration proposals.
Ingestion points: The script scripts/schema_health.py recursively reads .md files in the user-provided directory to extract frontmatter data.
Boundary markers: Frontmatter is delimited by standard --- markers, which the agent identifies during analysis.
Capability inventory: The skill uses scripts/schema_health.py for reading and the frontmatters CLI tool for modifying files (e.g., frontmatters set, frontmatters apply-base).
Sanitization: Data parsing is performed using yaml.safe_load(), which prevents most common deserialization attacks.
[COMMAND_EXECUTION]: The skill requires the agent to execute a local Python script and an external CLI utility to perform its core functions.
Evidence: Instructions guide the agent to run scripts/schema_health.py <dir> and use the frontmatters tool for migrations. Safety is mitigated by an explicit 'Never auto-apply' policy requiring human review of all proposed changes.

taxonomy-steward