The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script proposal.py uses importlib.util to dynamically load a local helper module from ../pipeline/deals.py. This is a standard modularity pattern used to share code between adjacent skills in the same environment using a hardcoded relative path.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from call logs and interaction history to draft proposals, creating a potential injection surface.
Ingestion points: Ingests data from closer/calls/ files and the output of the /record-keeper timeline command.
Boundary markers: The skill uses [Claude: ...] placeholders in the draft output to ensure the AI does not attempt to invent facts and to prompt human review.
Capability inventory: Operations are limited to local file system read/write within the closer/ directory. No network access or arbitrary system command execution capabilities are present.
Sanitization: The record command in proposal.py includes a validation check that prevents final logging if any [Claude: placeholders remain, enforcing a manual oversight step.

proposal-writer