The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the pipecat CLI and uv build tool to interact with cloud services and manage local environments. It also executes a local Python wrapper script scripts/safe_cloud_metadata.py to handle sensitive metadata operations.
[EXTERNAL_DOWNLOADS]: Instructions include fetching uv binaries from official sources, requiring the use of immutable digests for integrity verification during the preparation phase.
[DYNAMIC_EXECUTION]: Supports building Docker images and creating deployment archives from source code. The skill enforces strict staging procedures, such as verifying archive member hashes and preventing the use of symlinks, to ensure the integrity of the generated artifacts.
[DATA_EXFILTRATION]: Manages the upload of source code and configuration to Pipecat Cloud. To prevent accidental secret leakage, the skill explicitly excludes .env files, credentials, and virtual environments from the build context and utilizes a redaction boundary to protect the PIPECAT_TOKEN and Daily API key during metadata retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from cloud APIs and documentation which could potentially contain malicious instructions.
Ingestion points: Retrieves organization metadata, agent status, and CLI documentation via the pipecat tool.
Boundary markers: Employs a "value-safe redaction/allowlist boundary" for command outputs and documentation retrieval.
Capability inventory: Includes subprocess execution (pipecat), file system manipulation for staging, and network-based deployment operations.
Sanitization: Utilizes a specialized metadata wrapper script to emit only schema-validated, allowlisted metadata fields while withholding sensitive details like emails and credentials.

deploy