The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses terminal-based operations such as rm -rf for bulk removal of wiki directories and sha256sum for verifying the integrity of ingested source material.\n- [COMMAND_EXECUTION]: Utilizes python3 -c as a helper to parse JSON data retrieved from external sources, specifically during the processing of long-form articles from X (Twitter).\n- [EXTERNAL_DOWNLOADS]: Fetches content from external web URLs and social media platforms via tools like web_extract and xurl to populate the user's local wiki.\n- [EXTERNAL_DOWNLOADS]: Conducts freshness checks by pulling repository metadata and README files from GitHub to notify the user of updates to previously ingested sources.\n- [PROMPT_INJECTION]: The skill processes untrusted external data (web articles, X posts), creating a surface for indirect prompt injection. This risk is mitigated by the structured ingest workflow which focuses on synthesis and human-reviewed filing.\n
Ingestion points: External URLs, social media posts, and vault inbox files (as described in references/ingest-workflow.md).\n
Boundary markers: Missing specific instructions to ignore embedded commands within the ingested raw text.\n
Capability inventory: File read/write/delete access via TurboVault tools and general terminal execution for hashing and deletion.\n
Sanitization: Relies on the agent's summarization process; no programmatic filtering of the raw source content is performed before saving it to the vault.

vault-wiki