The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on potentially untrusted external data such as system logs, error messages, and documentation fetched via research tools. 1. Ingestion points: SKILL.md (processing error artifacts) and external-dependency-evidence.md (external documentation). 2. Boundary markers: The skill instructs the agent to use specific markdown formats for evidence. 3. Capability inventory: The skill has the ability to run shell commands, scripts, and utilize debugging tools (DAP, REPL). 4. Sanitization: The skill's core Iron Law mandates human authoritative confirmation for all causal propositions, providing a strong defense against automated exploitation via poisoned inputs.
[DYNAMIC_EXECUTION]: The skill incorporates the use of dynamic execution environments (debuggers, REPLs) as a primary method for hypothesis verification. This is a legitimate and intended feature of the debugging workflow.
[COMMAND_EXECUTION]: The agent is instructed to execute shell commands and test scripts to build a red-capable feedback loop for bug reproduction. These operations are intended for validation within local or dedicated environments.

root-cause