The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly reads from ~/Mundi Princeps/config/api_keys.json, which contains credentials for LSEG, Orbis, and A-Leads. Exposing raw secret files to the agent context is a high-risk pattern for accidental disclosure or misuse.
[COMMAND_EXECUTION]: The workflow involves executing local shell scripts, specifically apps/pitchbook-mcp/scripts/refresh-pb-cookies.sh. This execution occurs in an environment where the agent also has Write and Edit permissions, meaning the agent could potentially modify these scripts before execution.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its pipeline design.
Ingestion points: Untrusted data is ingested via mcp__perplexity__perplexity_research (web grounding) and external financial APIs (PitchBook, CapIQ).
Boundary markers: None are defined; there are no instructions for the agent to ignore or delimit embedded commands within the fetched dossiers or news articles.
Capability inventory: The skill possesses powerful capabilities including Bash access, file system Write/Edit, and arbitrary SQL execution via mcp__supabase__execute_sql.
Sanitization: There is no evidence of sanitization, filtering, or validation of the external content before it is processed or written to the final dossiers.

mundi-orch-counterparty-enrich