The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using subprocesses in 'scripts/bootstrap.py', 'scripts/speak.py', and 'scripts/publish.py'. These commands are used to manage the Python virtual environment, verify system dependencies like 'ffmpeg' and 'gh', and perform Git operations (clone, commit, push) for hosting the podcast feed on GitHub Pages. All executions are scoped to the tool's installation and data directories.
[EXTERNAL_DOWNLOADS]: During setup, the skill installs well-known Python dependencies (httpx, pydub, PyYAML, etc.) from the official PyPI registry via pip. It also uses the GitHub CLI to clone repositories for feed management. These actions are transparently documented in the installation instructions and are necessary for the skill's functionality.
[DATA_EXFILTRATION]: The skill transmits user-provided script text to the Fish Audio REST API ('api.fish.audio') to generate speech audio. This network communication is required for the tool's primary purpose and uses an API key provided by the user and stored in a local '.env' file.
[SAFE]: Indirect Prompt Injection Surface analysis:
Ingestion points: The skill ingests user-provided text, local files, and remote URLs to generate podcast scripts (SKILL.md, Generating an episode).
Boundary markers: While the skill processes external content, it relies on the underlying agent's instructions and 'PROMPT.md' to guide script generation rather than specific data delimiters.
Capability inventory: The skill has the capability to perform network requests (Fish Audio API) and execute shell commands (ffmpeg, git, gh) across its scripts.
Sanitization: No explicit sanitization or escaping of ingested content is performed before it is passed to the agent's context. However, this surface is typical for content-processing skills and is mediated by the agent's internal safety filters.

podcast