The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts and shell commands to automate repository tasks. Evidence includes the invocation of tools/lint-number-provenance.py, tools/lint-unsourced-confidence.py, and tools/source-cache.py, as well as git operations like pull, add, and merge.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch research data from external URLs using the tools/source-cache.py utility. This is a primary function of the research phase (Phase 0-2).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the internet during the research phase. 1. Ingestion points: External URLs fetched via tools/source-cache.py in Phase 0-2. 2. Boundary markers: The skill relies on natural language instructions to only use verified content but does not define strict data delimiters in the instructions. 3. Capability inventory: The agent has the ability to execute repository-local scripts and perform Git management tasks. 4. Sanitization: The skill implements a four-lens review process (Phase 4-5) that requires agents to verify all claims against source material before they are applied to the repository.

build-bundle