The Agent Skills Directory

Indirect Prompt Injection Risk: The skill processes external, untrusted content (inbound demand letters) which is then used to influence triage decisions and automated handoffs. There is a potential for instructions hidden within these letters to influence the agent's behavior.
Ingestion Points: The skill reads an incoming document from a user-provided path in Step 1.
Boundary Markers: The instructions do not define specific delimiters or "ignore embedded instructions" markers for the content of the inbound document.
Capability Inventory: The skill writes triage reports to the file system, updates a portfolio log (_log.yaml), and triggers handoffs to other functional skills (matter-intake, demand-intake).
Sanitization: The instructions do not specify sanitization or validation of the text extracted from the external demand letter.
Access to Sensitive Information: The skill accesses internal configuration and log files located in ~/.claude/plugins/config/claude-for-legal/. These files contain portfolio data, risk calibration profiles, and historical matter logs, which are highly sensitive in a legal context. While necessary for the skill's operation, this access creates a high-value target for information gathering.
Legal Safety Guardrails: The skill implements a "No silent supplement" policy and a "Source attribution" requirement. These features explicitly prevent the agent from autonomously generating legal research or filling gaps with model knowledge, instead requiring human verification of all citations. This is a positive safety pattern for reducing the risk of misinformation in specialized workflows.

demand-received