The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides examples for using the seamless-run CLI tool to execute shell commands as cached transformations (e.g., seamless-run 'seq 1 10 | tac && sleep 5'). These are presented as standard usage patterns for wrapping existing pipelines.
[REMOTE_CODE_EXECUTION]: The documentation describes several remote execution backends, such as jobserver and daskserver, which allow code to be executed on worker pools or HPC clusters. The skill also includes 'Remote execution don'ts' that advise against non-reproducible and insecure methods like ad-hoc script copying.
[PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection, as it is designed to analyze and process existing user-provided codebases and pipelines.
Ingestion points: The skill analyzes user-provided Python, bash, or hybrid pipelines (SKILL.md).
Boundary markers: The guidance emphasizes the use of explicit inputs and outputs and content-addressed I/O to maintain boundaries.
Capability inventory: The agent is given instructions on how to execute shell commands (via seamless-run) and Python functions (via @direct/@delayed decorators) extracted from the analyzed pipelines.
Sanitization: The skill relies on content-addressed checksums to ensure identity and integrity but does not specify logic-level sanitization for the input scripts.

seamless-adoption