The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends the installation of the @annals/agent-network package via pnpm. This is a vendor-owned package provided by the skill author for interacting with the A2A platform.\n- [COMMAND_EXECUTION]: The skill is based on the use of the ah CLI tool, which performs network operations, file system reads and writes, and local agent management tasks.\n- [DATA_EXFILTRATION]: The ah call and ah chat commands provide mechanisms for sending local files and text context to external agents using the --upload-file and --input-file flags. This creates a data exposure surface when interacting with third-party agents on the open platform.\n- [PROMPT_INJECTION]: The skill demonstrates a workflow for chaining multiple agents together, which introduces an indirect prompt injection surface.\n
Ingestion points: Data enters the context via the output of ah call tasks, which is then reused in subsequent commands (documented in SKILL.md Step 4).\n
Boundary markers: The provided examples do not implement delimiters or safety instructions to prevent the agent from obeying instructions embedded in the external output.\n
Capability inventory: The ah CLI possesses the capability to perform network requests (ah call), read files (--upload-file), and write files (--output-file).\n
Sanitization: The instructions show direct string interpolation of external agent output (e.g., ${TREND}) into subsequent task descriptions without sanitization or validation.

ah-a2a