The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes Axiom audit logs and query history which contain user-controllable data such as query strings and dataset names. This content is parsed by scripts/analyze-query-coverage and scripts/baseline-stats to generate cost-saving recommendations and reports, potentially allowing malicious query metadata to influence the analysis results.
Ingestion points: Audit data and query history are fetched from Axiom datasets in scripts/analyze-query-coverage and scripts/baseline-stats.
Boundary markers: No specific delimiters or "ignore instructions" warnings are used when processing the text of queries retrieved from history logs.
Capability inventory: The skill performs network operations via the Axiom API, writes to temporary files, and executes local shell scripts for data processing.
Sanitization: While the skill uses jq for handling JSON, it relies on string interpolation to build new APL queries and uses sed for dashboard template modification, which lacks robust escaping for the interpolated values.
[COMMAND_EXECUTION]: Several bash scripts (e.g., scripts/deploy-dashboard, scripts/create-monitors) construct APL queries and modify dashboard templates using string interpolation and sed with variables derived from user input or remote API responses (such as dataset names). While these commands primarily target the Axiom API, the pattern of using sed for template substitution is susceptible to injection if inputs contain delimiter characters or malicious command sequences.

controlling-costs