The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines autonomous triggers to initiate reasoning workflows and processes external data from code and documentation. Ingestion points include local codebase files and documentation. No explicit boundary markers are defined to isolate untrusted data from instruction processing. The agent has capabilities for file access and command execution, relying on internal refutation steps rather than programmatic sanitization.
[COMMAND_EXECUTION]: The grounding procedure involves running 'probes' or 'spikes', which include writing and executing temporary benchmarks or scripts to validate architectural decisions and hypotheses.
[DATA_EXFILTRATION]: The agent is instructed to read local codebase files and documentation to justify its choices. This creates a data access surface, though no evidence of unauthorized exfiltration or remote network transmission was detected.

free-will