The Agent Skills Directory

[PROMPT_INJECTION]: The skill reads failure logs from kb/learnings/ and uses identified patterns to modify agent system prompts in app/agents/*.md. This creates an attack surface for indirect prompt injection, as malicious instructions within logs could be adopted into the agent's core instructions.\n
Ingestion points: Reads failure patterns and log entries from files in kb/learnings/ and existing agent definitions in app/agents/.\n
Boundary markers: No specific delimiters or instructions are provided to the agent to treat input from logs as untrusted data or to ignore embedded instructions.\n
Capability inventory: Employs Read, Grep, and Edit tools to process data and rewrite prompt/skill files; calls scripts/validate.py via subprocess.\n
Sanitization: The skill mandates a check for forbidden code patterns (eval, exec, shell=True) after editing files, but does not provide sanitization for natural language instructions being interpolated into prompts.\n- [COMMAND_EXECUTION]: The skill invokes python3 scripts/validate.py to check the structural integrity and frontmatter validity of modified files.

evolve