The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs network requests to external APIs to retrieve exercise and nutrition data.
Targets include the wger open database (wger.de) and the official USDA FoodData Central API (api.nal.usda.gov).
These requests are directed to well-known and reputable services for health and nutrition information.
[REMOTE_CODE_EXECUTION]: Automated scanners flagged several patterns involving curl piped to python as potential remote code execution.
Review of the code in SKILL.md shows that these are pipes to inline Python scripts (using python -c) defined within the skill itself to process JSON data from the APIs.
This pattern executes local logic on remote data and does not constitute the execution of untrusted remote code.
[COMMAND_EXECUTION]: The skill uses shell commands (curl, python) to interact with APIs and run utility scripts.
User-provided search queries (e.g., $QUERY, $FOOD) are correctly sanitized using urllib.parse.quote via a Python subshell to prevent command injection in URLs.
Some procedural examples in SKILL.md (e.g., using $EXERCISE_ID or $FILTER) do not explicitly apply this encoding, which could potentially allow for argument injection if an agent interpolates malicious input into the shell command.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external APIs, which represents a potential surface for indirect prompt injection if those sources were compromised.
Ingestion points: SKILL.md fetches and displays exercise and food descriptions from wger and USDA APIs.
Boundary markers: Delimiters or explicit instructions to ignore embedded commands are absent.
Capability inventory: The skill possesses network and shell command execution capabilities via curl and python subprocesses.
Sanitization: The skill employs re.sub('<[^>]+>','',html.unescape(...)) to strip HTML tags from exercise descriptions, providing a basic defense against malicious scripts or markup in the API responses.

fitness-nutrition