The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: In references/installation.md, the skill directs users to install the tool by downloading a script from https://cli.momentic.ai/qa and piping it directly into the shell (sh). This pattern executes remote code without integrity verification such as checksums.
[EXTERNAL_DOWNLOADS]: The references/installation.md file instructs the installation of a global Node.js package named qa (npm install -g qa@latest). The use of such a generic name for a package poses a supply chain risk, as it may conflict with other packages or be subject to name shadowing.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection as it ingests findings from a remote agent to perform automated repairs on local code.
Ingestion points: The skill reads report.json, bugs.json, and verdicts.json via the qa report and qa status commands in references/reports.md and SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content within the reports.
Capability inventory: The skill has the capability to execute shell commands and modify local source code based on these reports during the 'Repair loop' described in references/remediation-loop.md.
Sanitization: The skill lacks automated sanitization for the content of the findings, relying instead on the agent to manually inspect them.
[COMMAND_EXECUTION]: The skill extensively uses shell commands to interact with the qa CLI. This includes starting tunnels to expose local ports in references/tunneling.md and executing a complex while-loop for status monitoring in SKILL.md.

mo-qa