The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that involves ingesting large amounts of untrusted data from external sources, such as annual reports, 10-Ks, and earnings call transcripts (Stage 1 in SKILL.md and references/01-data-sourcing.md). This ingestion surface could potentially be exploited by indirect prompt injection attacks if an attacker were to embed malicious instructions within these financial documents.
Ingestion points: Untrusted data enters the agent context through the acquisition of primary filings and transcripts as described in Stage 1 of the core workflow.
Boundary markers: The instructions do not specify the use of delimiters or explicit warnings to the model to ignore instructions found within analyzed documents, which could lead to accidental obedience of adversarial content.
Capability inventory: The agent has the capability to execute local utility scripts (scripts/ratios.py, scripts/score.py, scripts/valuation.py, scripts/verify_data.py, and scripts/lint_report.py) as part of its analysis process.
Sanitization: There is no evidence of explicit input sanitization or filtering of external content before it is processed by the agent or passed to the local scripts.

stock-analysis