The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local Python scripts to perform its operations, including list_new_notes.py, pull_transcript.py, export_thread_bodies.py, and score_run.py. These scripts are used to interface with external APIs and evaluate the success of the ingestion routine.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from various external communication channels and meeting transcripts, which are then used to update internal records and generate communication drafts.
Ingestion points: Meeting transcripts from Granola (pull_transcript.py), email thread archives from Gmail (export_thread_bodies.py), Slack message/thread history, and LinkedIn engagement data.
Boundary markers: The skill defines "three standing rails" that require manual verification for financial data and dates, and strictly prohibit automated sending of communications or bulk deletion of records.
Capability inventory: The skill can perform file writes to deal folders and dashboards, update CRM records (Attio), execute Git commits to the local repository, and create email drafts in Gmail.
Sanitization: No explicit sanitization or filtering of the ingested text is mentioned beyond the requirement to verify specific data points against primary sources.
[CREDENTIALS_UNSAFE]: The skill specifically checks for the presence of numerous sensitive API tokens and keys in the environment, including GRANOLA_API_KEY, ATTIO_API_KEY, APIFY_API_TOKEN, POSTBRIDGE_API_KEY, and several Slack/Gmail authentication tokens. While it does not hardcode these values, it manages an environment with high-value credentials.

recoup-internal-consulting-nightly-ingestion