skills
mbfinotti/revops-skills/skills
- Type
- Skill
- Scanned
- Nov 13, 2026
- Version
- 12fd9f5537bb
- Files read
- 118
This project is a collection of twenty Markdown-based skills that guide an AI agent through Revenue Operations workflows such as lead scoring, lead routing, pipeline hygiene, forecasting, deal desk approvals, and data governance. It contains no executable code of its own, and the main risk is minor workflow overreach where skills instruct the agent to edit project-wide agent-instruction files or install background web-fetching routines without explicit user approval.
What it can reach
- Files
- Instructs the agent to read project documentation, CSV exports, and agent-instruction files, write deliverable specs, update revops-context.md and persistent memory files, and optionally edit the project's agent-instruction file.
- Network
- The revops-radar skill instructs the agent to fetch external websites, blogs, newsletters, and public forums to verify whether recommended RevOps sources are still active.
- Shell commands
- Instructs the agent to read git log history and, when code execution is available, run local scripts to reconcile user-provided billing, CRM, and forecast export files.
- Other
- Instructs the agent to query connected CRM, billing, analytics, ticketing, or HR integrations when exposed by the host harness, and to propose or install scheduled routines.
Findings
-
LowKickoff skill permits modifying project agent-instruction files without user approval
skills/revops-kickoff/SKILL.md:190
The revops-kickoff skill is configured to run automatically at the start of any RevOps task, where it reads git history, project files, and connected ticketing, CRM, or meeting-note integrations. It then authorizes the agent to patch the project's root agent-instruction file (such as CLAUDE.md or AGENTS.md) without asking for confirmation, unlike its git-memory step which requires showing a diff first. Unapproved edits to agent-instruction files affect every future session in the repository and could allow prompt-injected text from external tickets or notes to persist.
Fix: Require the agent to show the proposed diff and obtain explicit user approval before modifying any agent-instruction file, matching the approval rule already used for git-backed memory.
-
LowRadar skill installs a background web-fetching routine without confirmation
skills/revops-radar/SKILL.md:127
Unlike revops-kickoff, which requires a dry run and explicit user approval before scheduling any recurring task, revops-radar tells the agent to directly install a quarterly refresh routine whenever the host harness supports scheduled routines. Each refresh fetches external URLs—including public forums like Reddit and domains that may have expired and been resold to strangers—exposing unattended background agent runs to untrusted web content without interactive oversight.
Fix: Require explicit user confirmation before creating the scheduled refresh routine, and instruct the agent to treat all fetched web page content strictly as untrusted data.
