The Agent Skills Directory

[SAFE]: The skill implements strict read-only enforcement in the 'query' mode by using MATERIALIZED transaction isolation and connection-level blocks, ensuring that INSERT, UPDATE, DELETE, and MERGE operations are rejected even if requested by a user.
[SAFE]: Advanced credential protection is present across all modules; the skill is designed to discover configurations in secure paths (e.g., .env/databases/) and explicitly redacts secrets from logs, results, and error messages using a [REDACTED] mask.
[SAFE]: The migration-guard.ts script provides legitimate security utility by auditing SQL migrations for high-severity locks (Level 8 ACCESS_EXCLUSIVE), preventing table rewrites, and identifying unindexed foreign key hazards that could cause production lockouts.
[SAFE]: Execution guardrails such as a 30-second timeout, 10,000-row caps, and 100-character column limits prevent runaway queries from exhausting database resources or overwhelming the agent's context.
[INDIRECT_PROMPT_INJECTION]: The skill has a standard attack surface for database tools as it ingests untrusted data from query results and external SQL/Markdown files. This is well-mitigated by the skill's defense-in-depth architecture, which treats all external data as non-executable and forces structured ASCII table formatting for results.
[SAFE]: Usage of the bun runtime for helper scripts (cognitive-graph.ts and migration-guard.ts) is documented for local execution on the operator's machine and does not involve downloading or executing remote code from untrusted sources.

database