The Agent Skills Directory

Internal Command Execution: The skill is configured to execute local Python scripts, specifically user_context_preflight.py and validate_payload.py, to manage session context and validate data integrity. These operations are restricted to internal skill paths and are used for operational consistency.
Indirect Prompt Injection Surface: The skill ingests untrusted external data from sources like company filings, earnings transcripts, and web search results. This creates a potential surface for indirect prompt injection, where malicious instructions embedded in documents could attempt to influence the agent's behavior.
Ingestion points: Data enters the context via company_filings_ir, earnings_transcripts_presentations, and web search connectors described in SKILL.md and references/source-policy.md.
Boundary markers: The skill utilizes references/source-policy.md to define source priority and facts vs. assumptions, though explicit 'ignore instruction' delimiters for raw content are not defined in the provided files.
Capability inventory: The agent can execute shell commands (python scripts), generate HTML artifacts, and interact with various financial support tools.
Sanitization: The workflow incorporates a quality check via validate_payload.py and the references/quality-workflow.md to ensure evidence is sourced and numerically auditable.

memo-builder