The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates database queries by executing psql and mysql CLI tools through a wrapper script scripts/db_readonly.sh. This script uses shell arrays to safely pass arguments, preventing shell command injection, and includes a regex filter to block destructive SQL statements such as DROP, DELETE, and ALTER.
[DATA_EXFILTRATION]: The execution script includes an --out parameter that allows query results to be written to local files. To prevent unauthorized data transfer, the skill specifies clear instructions that prohibit sending results to external network endpoints or third-party services.
[SAFE]: The skill implements advanced safety controls documented in SKILL.md and SAFETY.md, specifically blocking EXPLAIN ANALYZE commands that could bypass read-only constraints and mandating LIMIT clauses on exploratory queries to prevent resource exhaustion.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via database content.
Ingestion points: Results from SQL queries executed via scripts/db_readonly.sh.
Boundary markers: Lacks explicit delimiters for separating query data from instructions, although prompt guardrails advise on safe handling of results.
Capability inventory: Execution of shell-based database clients and local file writing capabilities in scripts/db_readonly.sh.
Sanitization: SQL inputs are filtered via a keyword-based regex in the shell script, but the content of the data retrieved from the database is not sanitized before it is provided to the agent context.

db-readonly-hardened