The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is explicitly designed to identify and process sensitive business and financial data from the local environment.
Evidence: Phase 1.1 directs the agent to search for 'Financial reporting submissions: P&L, balance sheet', 'Employee satisfaction / engagement survey data', and 'POS system aggregate data across locations'. Accessing these file types constitutes significant exposure of private organizational information.
[PERSISTENCE]: The skill includes instructions to maintain state and record its own execution metadata across sessions, which is a hallmark of persistence.
Evidence: The 'SELF-EVOLUTION TELEMETRY' section instructs the agent to check for project memory directories at ~/.claude/projects/ and append execution logs to a file named skill-telemetry.md. This allows the skill to influence future iterations of itself or other agent actions through stored state.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted third-party data which could contain malicious instructions designed to influence the agent's analysis or reporting.
Ingestion points: Phase 5.1 and 5.2 involve aggregating 'guest comment analysis', 'online review aggregation' (Google, Yelp), and 'Employee satisfaction / engagement survey data'.
Boundary markers: There are no instructions providing delimiters or warnings to ignore embedded directives within these external datasets.
Capability inventory: The skill has the capability to read extensive local file structures (Phase 1) and write files to the local disk (Phase 6 and Telemetry section).
Sanitization: No sanitization or filtering logic is specified for the external review or survey content before it is incorporated into the final report or telemetry logs.

franchise-benchmarking