The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface where untrusted data is processed without sufficient safeguards.
Ingestion points: Untrusted error messages and generation traces are fetched from the database using SQL queries in the 'Step 1: Fetch New Errors' section of SKILL.md.
Boundary markers: The instructions lack the use of delimiters or specific boundary markers when interpolating raw error data into the Task delegation tool or bd CLI commands.
Capability inventory: The skill has powerful capabilities, including database modification (mcp__supabase__execute_sql), task management through the bd CLI, shell command execution (pnpm), and delegation to subagents for automated code changes.
Sanitization: There is no evidence of sanitization or escaping applied to the error_message or error_data content before it is processed by the agent or passed to other tools.
[COMMAND_EXECUTION]: The skill relies on the execution of local shell commands using the bd (Beads) CLI tool for creating and tracking bug tasks, as well as pnpm for running type checks and project builds during the verification phase.

process-logs