The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by incorporating task-derived evidence into an HTML file that is subsequently executed in the user's browser.
Ingestion points: Task findings, metrics, and transcript evidence are processed in SKILL.md (Workflow Step 3) to build the self-contained report.
Boundary markers: Absent. The instructions do not specify any delimiters or warnings to ignore instructions embedded within the ingested data.
Capability inventory: The skill has the ability to execute shell scripts (prepare.sh), write to the filesystem (mkdir), and launch external applications (open).
Sanitization: There is no evidence of HTML escaping or sanitization of the evidence data before it is written to the report, which could allow malicious scripts to execute when the file is opened.
[COMMAND_EXECUTION]: The skill invokes a local bash script scripts/prepare.sh to set up the environment and uses the open command to automatically launch the generated HTML report in the default browser. The prepare.sh script includes robust regex validation for the SLUG argument, which mitigates standard command injection and path traversal risks.
[EXTERNAL_DOWNLOADS]: The skill identifies a dependency on the html-playground skill and provides the user with an installation command using bunx that targets a repository owned by the author (PaulRBerg/agent-skills). This represents an intended use of vendor-controlled resources.

html-debrief