The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Ruby script (scripts/graph-audit) as a final step in the setup and migration processes to ensure vault consistency.
Evidence: SKILL.md Step 6: "Run scripts/graph-audit across the complete fresh or migrated vault."
[PRIVILEGE_ESCALATION]: The agent is directed to create a script file in the user's environment and explicitly modify its file permissions to be executable.
Evidence: SKILL.md Step 5: "Copy references/graph-audit into it byte-for-byte as scripts/graph-audit and preserve executable mode (or set it where the filesystem supports executable bits)."
[INDIRECT_PROMPT_INJECTION]: The skill defines a standard architecture for ingesting data from an external "inbox" directory, creating a surface where malicious instructions embedded in documents could influence the agent's behavior.
Ingestion points: The raw/inbox/ directory, intended for Articles, transcripts, and exports, as specified in references/vault-scaffold.md.
Boundary markers: The system mandates the use of Founder-stated: unverified labels and [unverified] markers in content bodies to distinguish between stated and verified data, as detailed in references/CONVENTIONS-core.md.
Capability inventory: The skill possesses the capability to read/write files throughout the vault structure and execute shell commands via the graph-audit script.
Sanitization: Content is fingerprinted using SHA-256 for idempotency, but there is no explicit validation of content safety before it is added to the agent's context, as noted in references/CONVENTIONS-inbox-ledger.md.

co-founder-setup