The Agent Skills Directory

Indirect Prompt Injection Surface: The skill ingests data from external sources and systems during its spike tasks to inform reasoning and update the assumption ledger. There is a potential consideration where instructions embedded in these external sources could influence the agent's behavior. The skill partially addresses this by emphasizing primary evidence and using a signed confidence scoring system.
Dynamic Execution of Spikes: The core functionality involves running 'spikes,' which include hands-on trials, empirical investigations, or tool executions. The skill provides instructions for isolating these executions using git worktrees or temporary directories to prevent cross-contamination or impact on the main project.
Command Execution for Ledger Updates: The skill uses a Python script, update_confidence.py, to programmatically update the risky-assumptions.yaml file. This script acts as a guardrail by ensuring that only specific fields (confidence and derisking) are modified, protecting other metadata from unauthorized changes.
Dynamic Loading in Tests: The included test suite (test_update_confidence.py) utilizes importlib.util to dynamically load the ledger update script by its filesystem path. This is a common practice for testing standalone scripts but involves loading executable code from a calculated path.

derisk-assumptions