The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted Terraform files and configuration data defined in the paths section of SKILL.md (e.g., _modules/**/*.tf). Boundary markers are present in SKILL.md under the 'Reviewing untrusted input' section, which explicitly instructs the agent to treat reviewed content as data and ignore embedded instructions. The capability inventory includes the Bash tool for local sed edits and the Read tool for file access. Sanitization is primarily handled through the agent's instructions to maintain role boundaries and treat input as non-authoritative data.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform targeted patching of local Terraform module source code using sed as described in SKILL.md. This operation is intended to resolve known upstream bugs in the CloudDrove module ecosystem and operates within the .terraform directory.

wrapper-tf