Use when a saved plan may contain hidden assumptions disguised as facts, especially for medium or large tasks, unresolved high-risk plans, or plans that mention files, APIs, versions, integrations, tests, or concurrency patterns that should be verified against the repository before execution.
Installs into .claude/skills of the current project.
Are you the author of Controlflow Assumption Verifier?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/smithbox-ai-controlflow-assumption-verifier)
---
name: controlflow-assumption-verifier
description: "Use when a saved plan may contain hidden assumptions disguised as facts, especially for medium or large tasks, unresolved high-risk plans, or plans that mention files, APIs, versions, integrations, tests, or concurrency patterns that should be verified against the repository before execution."
---
# ControlFlow Assumption Verifier
## Overview
Hunt mirages in a saved plan. This skill adapts ControlFlow's AssumptionVerifier to Codex by checking whether the plan invents files, APIs, dependencies, conventions, or execution assumptions that the repository does not support.
## Workflow
1. Read the saved plan artifact first.
2. For each important claim in the plan:
- identify the claim
- decide whether it is repository-verifiable
- verify it against files, paths, symbols, configs, or tests
- classify it as `VERIFIED`, `UNVERIFIED`, or `MIRAGE`
3. Use [references/mirage-patterns.md](references/mirage-patterns.md) as the review grid.
4. Focus especially on:
- phantom paths
- phantom APIs
- missing dependencies
- pattern mismatches
- missing error paths
- missing migrations
- missing security boundaries
5. Save the verdict to `plans/artifacts/<task-slug>/assumption-verifier.md` using `../../templates/assumption-verifier-report-template.md`.
6. Return a structured text report with:
- `COMPLETE` or `ABSTAIN`
- blocking and minor mirages
- a dimensional scorecard
- a recommendation
7. If the plan requires meaningful redesign, classify it as:
- `fixable`
- `needs_replan`
- `escalate`
## Output Shape
- **Status**
- **Mirages Found**
- **Dimensional Scores**
- **Summary**
- **Failure Classification** when blocking issues exist
## Common Mistakes
- Treating every unknown as a mirage instead of separating `UNVERIFIED` from `MIRAGE`.
- Checking only presence claims and skipping missing requirements or edge cases.
- Reviewing the plan abstractly without actually opening the referenced files or paths.
## References
- `references/mirage-patterns.md`
- `../../templates/assumption-verifier-report-template.md`