The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install a CLI utility from '@mvanhorn/printing-press-library' on NPM and 'github.com/mvanhorn/printing-press-library' on GitHub. These sources align with the vendor identity established in the skill metadata.
[REMOTE_CODE_EXECUTION]: Installation steps involve using 'npx' and 'go install' to fetch and execute binary code directly from remote vendor-controlled repositories.
[COMMAND_EXECUTION]: The skill operates by executing shell commands via the 'averusa-pp-cli' binary to perform document harvesting, searching, and spec comparisons.
[DATA_EXFILTRATION]: The CLI tool includes a '--deliver webhook:' feature that allows routing command output (including document text and metadata) to arbitrary external HTTP endpoints via POST requests. This represents a potential egress path for data retrieved by the agent.
[PROMPT_INJECTION]: The skill implements an automated 'learning loop' and 'playbook' system that stores instructions in a local SQLite database to influence future agent behavior.
Ingestion points: The 'averusa-pp-cli recall' command reads previously stored instructions from the local database ('data.db'), and 'averusa-pp-cli harvest' ingests data from the AVer support portal.
Boundary markers: The skill does not define specific delimiters or instructions to prevent the agent from following malicious commands embedded in harvested documents or stored learnings.
Capability inventory: The skill has access to the 'Bash' tool and can perform file system operations and network requests through the CLI.
Sanitization: There is no automated sanitization described for the content stored in the learning loop database; the agent is only manually instructed to strip PII before teaching.

pp-averusa