The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data that could influence the AI agent's behavior during report analysis.
Ingestion points: Recipient names and arbitrary note strings are provided via the mint command (in scripts/decklinks.mjs) or the admin API (in app/server.mjs) and stored in the database.
Boundary markers: The report command in the CLI outputs these strings with standard labels (e.g., "Recipient · deck · URL"), but lacks specific delimiters or warnings to the agent to ignore instructions embedded within these fields.
Capability inventory: The agent has capabilities to execute shell commands, publish files, manage the server, and mint new links based on the instructions it receives.
Sanitization: While the web-based dashboard (app/admin.html) correctly escapes HTML entities, the plain-text reports consumed by the AI agent do not have specific sanitization or filtering to prevent prompt injection attacks embedded in recipient names or notes.
[COMMAND_EXECUTION]: The skill frequently executes local shell commands to manage its lifecycle and process files.
It uses node:child_process to spawn its own server in the background and to run system utilities like pdftoppm and pdftotext for PDF-to-HTML conversion.
These operations are central to the skill's primary purpose and use static or strictly validated command names.
[EXTERNAL_DOWNLOADS]: The skill documentation and doctor command check for and advise the installation of external system dependencies.
It requires poppler-utils (specifically pdftoppm) for PDF deck processing.
It also supports optional push notifications via the ntfy.sh service, which involves sending data to an external, well-known service.

tracked-deck-links