The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and analyze untrusted codebase content. While it uses a senior security engineer persona and evidence-anchoring rules to maintain objectivity, the processing of external data (source code) presents a standard surface for indirect prompt injection.
Ingestion points: Reads and summarizes codebase files from user-specified paths as defined in references/prompt-template.md.
Boundary markers: The skill defines a structured output contract and senior engineer persona but does not implement explicit delimiters to isolate untrusted file content within the prompt.
Capability inventory: Uses standard search tools (ripgrep) and generates a markdown report; no high-risk capabilities like network access or arbitrary code execution are present.
Sanitization: The skill instructions include a mandatory rule to redact secrets found in the code, though it does not specify character-level sanitization for ingested text.
[COMMAND_EXECUTION]: Developer Tool Integration. The skill prompts the agent to utilize standard developer tools like ripgrep (rg) for exploring codebases.
Evidence: Instructions in references/prompt-template.md suggest using rg -I to locate security-relevant code patterns. This is an appropriate and expected use of command-line tools for a threat modeling task.

security-threat-model