The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill contains a script scripts/sync-upstream-harness.sh that fetches numerous markdown files and templates from an untrusted external repository at raw.githubusercontent.com/revfactory/harness/main. While these files are documentation, they serve as the ground truth for the skill's generation logic.
[REMOTE_CODE_EXECUTION]: The provided scripts/install.sh script instructs the system to download and install the skill from the external revfactory/harness repository directly into the global skill configuration directory (~/.claude/skills).
[COMMAND_EXECUTION]: Several bundled shell scripts, including bootstrap-harness.sh and validate-harness.sh, are executed to perform project scaffolding and integrity checks on generated agent definitions.
[PROMPT_INJECTION]: The skill has a significant vulnerability surface for indirect prompt injection. It is designed to read and analyze user codebases to identify domains and task types (documented in SKILL.md Step 1). Malicious instructions embedded in the analyzed codebase could influence the generation of agent roles and orchestration logic.
Ingestion points: SKILL.md (Step 1: Domain Analysis) specifies reading the codebase or user request to identify roles.
Boundary markers: Absent. There are no instructions to use delimiters or ignore instructions found within the analyzed data.
Capability inventory: The skill utilizes powerful tools including Bash, Write, Edit, Agent, and TeamCreate across its operation and the templates it generates.
Sanitization: Absent. No filtering or validation of the content extracted from the analyzed codebase is performed before it is interpolated into agent definitions.

harness