The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to operate in 'AUTONOMOUS MODE' and explicitly states 'Do NOT ask questions. Detect everything from the codebase and proceed.' This overrides standard agent interaction protocols and safety filters that require user confirmation for high-risk actions.
[INDIRECT_PROMPT_INJECTION]: The skill reads and analyzes various project files (package.json, database models, API routes) to determine framework and data structures. It lacks boundary markers or instructions to treat these external files as untrusted data, making it vulnerable if the codebase contains malicious instructions intended to hijack the agent's logic.
Ingestion points: Project root files, package.json, database schemas, API routes, and the $ARGUMENTS variable.
Boundary markers: Absent. There are no instructions to delimit or ignore instructions found within the analyzed files.
Capability inventory: File system modification (.env, src/), package installation (npm, pip, go), and script execution (npx ts-node).
Sanitization: Absent. No escaping or validation is performed on the data read from the codebase.
[DATA_EXFILTRATION]: The 'Self-Evolution Telemetry' section automatically writes execution metadata to ~/.claude/projects/skill-telemetry.md. This represents a automated write of operational metadata to a hidden directory outside the immediate project workspace.
[COMMAND_EXECUTION]: The skill performs broad shell command execution, including package management (npm/yarn/pip/go install) and running dynamic reindexing scripts (npx ts-node).
[EXTERNAL_DOWNLOADS]: The skill downloads and installs various search SDKs. These target well-known and official repositories for Algolia, Meilisearch, Typesense, and Elasticsearch.

search