The Agent Skills Directory

[COMMAND_EXECUTION]: The skill enables the execution of SQL queries against local PostgreSQL and remote HTTPS SQL services. It proactively mitigates SQL injection risks by requiring the use of bound parameters (%s) and explicitly forbidding string interpolation in both its documentation and implementation.
[EXTERNAL_DOWNLOADS]: The scripts/remote.py utility communicates with an external HTTPS service to store and retrieve task traces. It maintains a high security posture by validating TLS certificates, refusing HTTP redirects, and requiring authentication via tokens stored in owner-restricted private files (mode 0600).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and display "evidence" and "memory" from database tables, which may contain untrusted data.
Ingestion points: SQL query results from catalog.section, catalog.source_file, and working.item are returned to the agent context via scripts/context.py and scripts/remote.py.
Boundary markers: The skill instructions explicitly warn the agent that retrieved memories are "untrusted contributions" and must be verified against source sections.
Capability inventory: The skill possesses database connectivity (psycopg), network request capabilities (urllib), and local file read/write access (restricted to the configured database/config paths).
Sanitization: The implementation uses strict JSON serialization for results and enforces SQL parameterization to ensure that data retrieved from the database cannot manipulate the query logic or the agent's control flow.

sql-context