The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the 'unslop' utility using npx skills add from skills.sh. This involves downloading and executing content from an external domain that is not recognized as a trusted organization or well-known service.
[COMMAND_EXECUTION]: The skill provides shell commands to extract sensitive API keys from local environment files (.env.local) and write them to temporary files ($TMPDIR/exa.hdr) for use in subsequent network requests. It also recommends manual package installations using pip install for the rembg library and several other dependencies to handle image processing.
[INDIRECT_PROMPT_INJECTION]: The workflow involves reading customer-provided data, such as chat message bodies and email history, which are then used to synthesize outreach drafts and meeting plans.
Ingestion points: The skill reads chat_messages bodies, email threads from the Resend log, and Attio CRM notes to understand lead context.
Boundary markers: The instructions explicitly mandate a human-in-the-loop review process (defined in references/outreach-approval.md) where every draft must be reviewed and approved by the operator before any email is sent.
Capability inventory: The skill has capabilities to perform network requests to various APIs (Recoup, Attio, Privy, Stripe, Exa, Resend), execute SQL queries on production databases via Supabase, and write files to a local workspace.
Sanitization: While the workflow includes an 'unslop' pass to strip machine-like writing styles, there is no mention of sanitizing or escaping external inputs for potential malicious instructions before processing.

recoup-internal-sales