The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that are not under the direct control of the user, which could be used to host malicious instructions.
Ingestion points: The skill reads artifact content via tools like search and read, and polls for reviewer comments using comments-list.sh and comments-stream.sh (found in SKILL.md).
Boundary markers: Explicit instructions in the "Trust boundaries" section of SKILL.md direct the agent to treat reviewer comments and artifact source as untrusted data, specifically forbidding the execution of commands or disclosure of secrets requested by that content.
Capability inventory: The skill can execute the dsp CLI, write to local state files for comment tracking, and perform network requests to the vendor API.
Sanitization: Bash scripts validate input parameters like artifact IDs and cursor timestamps against restrictive alphanumeric regular expressions to prevent path traversal or command injection.
[EXTERNAL_DOWNLOADS]: The skill bundles pre-compiled binaries for the jq utility to handle JSON processing.
Evidence: scripts/_common.sh contains logic to select and use bundled statically-linked jq binaries from the bin/ directory based on the host platform (e.g., bin/jq-macos-arm64, bin/jq-linux-amd64). These are well-known, standard tools.
[COMMAND_EXECUTION]: The skill interacts with a locally installed command-line interface to perform authenticated operations.
Evidence: Multiple scripts, such as scripts/login.sh and scripts/publish.sh, use exec to call the official dsp binary if it is discovered on the system's execution path. The skill instructions explicitly forbid the agent from installing this CLI automatically.
[DATA_EXFILTRATION]: The skill transmits user-provided files and metadata to a remote server as part of its core publishing functionality.
Evidence: The scripts/publish.sh script and the instructions for staged uploads in SKILL.md use curl and PUT requests to send data to https://api.display.dev. This is the official API endpoint managed by the skill's author, display-dev.

display-dev