The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs dynamic loading and execution of Python functions based on string inputs in references/job-dispatch.md and references/job_runner.py.
It utilizes importlib.import_module(module) and getattr(mod, job_name) to resolve and execute code at runtime.
[COMMAND_EXECUTION]: The skill provides logic to spawn subprocesses for local task execution in references/job-dispatch.md.
The dispatch_job function uses asyncio.create_subprocess_exec to run tasks as independent Python processes when configured in subprocess mode.
[COMMAND_EXECUTION]: The skill documents the use of shell pipes and environment variables for database management tasks in SKILL.md and references/schema.md.
Example provided: procrastinate schema --print-schema | psql $DATABASE_URL.
[DATA_EXFILTRATION]: The skill reads captured command output from temporary files on the local file system in references/job-dispatch.md.
The get_job_history function uses log_path.read_text() to retrieve data stored in the system's temporary directory during subprocess execution.
[PROMPT_INJECTION]: The design of the job dispatching mechanism presents a surface for indirect prompt injection.
Ingestion points: Task names (job_name) and module paths (module) are accepted as string parameters in the dispatch_job function within references/job-dispatch.md.
Boundary markers: No explicit boundary markers or instructions to ignore instructions embedded in the task data are implemented.
Capability inventory: The skill possesses capabilities for subprocess execution and dynamic loading of Python modules.
Sanitization: No sanitization, validation, or allow-listing is performed on the module or task identifiers before they are used to load and run code.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to download and install the Procrastinate library from external package registries.
SKILL.md references standard installation commands like pip install procrastinate and pip install 'procrastinate[django]'.

procrastinate