The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill executes a local Python script (scripts/check_resignation_plan.py) to calculate notice periods, weekday gaps, and paid leave eligibility. This execution is limited to a static script provided within the skill's own directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data in JSON format to generate reports. An analysis of the attack surface reveals robust mitigations:
Ingestion points: Data enters through a JSON payload described in references/procedure.md and is processed by the check script.
Boundary markers: The instructions explicitly require the agent to leave unverified fields as unknown and prevent the agent from providing definitive legal or financial conclusions.
Capability inventory: Capabilities are restricted to local script execution via python3. There are no network operations, file writing capabilities (beyond standard output), or subprocess calls to untrusted binaries.
Sanitization: Input validation is performed by scripts/_common.py, which enforces strict data types for dates, integers, and specific enumerated choices. It also contains regex patterns (e.g., MYNUMBER_PATTERN) used to identify and potentially guard against the inclusion of sensitive Japanese personal identifiers.
[DATA_EXFILTRATION]: The skill instructions and associated documentation (SKILL.md, references/procedure.md) contain explicit prohibitions against exfiltrating business data or technical information to new employers. It also specifically directs the agent not to request or store sensitive credentials like MyNumber or pension identifiers.

resignation-plan