| name | verification-loop |
| description | A comprehensive verification system for Claude Code sessions. |
| origin | ECC |
Verification Loop Skill
A comprehensive verification system for AmdocsBrain sessions.
When to Use
Invoke this skill:
- After completing a feature or significant code change
- Before creating a PR
- Before marking a Todoist task complete
- Before reporting that work, a brief, or an operational artifact is complete
- After refactoring
Verification Modes
Choose the modes that match the work. Not every session needs every mode.
Mode 1: Universal Completion Gate
Run this for every completion claim:
- Check the task or request's success criteria against the actual output.
- Record the artifact path, link, or exact location.
- If the work is tracked in Todoist, confirm the live task state.
- Capture blockers, follow-ups, and residual risks.
- State evidence confidence as
verified, partial, or unavailable.
If any item is missing, the work is not ready to be described as complete.
Mode 2: Operational Verification
Use this for daily briefs, snapshots, reports, triage, and other workflow
artifacts.
Verify:
- the live sources you relied on were actually queried in this session
- degraded or unavailable sources are called out explicitly
- blockers and waiting-for items came from the latest trusted source
- you separated fact from inference
- the artifact links to the underlying evidence where possible
- the artifact meets the operational snapshot contract when that output is in scope
Mode 3: Engineering Verification
Use this when files, scripts, or code changed.
Run the repo-relevant checks that apply:
- build
- type check
- lint
- tests
- changed-file review
- targeted security / secret scan
Prefer workspace tools and repo-native commands over generic examples.
If a check does not apply, mark it N/A instead of silently skipping it.
Contract Links
- Todoist lifecycle and close-out semantics:
Core/10_TODOIST_TEAM_WORKFLOW.md
- Workspace completion gate:
Agents/WORKSPACE-AGENT-OVERLAY.md
- Operational snapshot contract:
Tools and Skills/Operational-Snapshot/README.md
Output Format
After verification, produce a concise report:
## Verification Report
- Scope: [what was verified]
- Success criteria: [pass/fail or short note]
- Artifact path: [path or N/A]
- Todoist state: [verified / N/A / issue]
- Source health: [verified / partial / unavailable]
- Engineering checks: [list pass/fail/N/A]
- Residual risks: [none or short list]
- Overall: [ready / not ready]
Continuous Mode
For long sessions, run verification after each meaningful milestone:
Set a mental checkpoint:
- After finishing a task
- Before moving a Todoist item to closed
- Before reporting completion to the user
- Before publishing a brief, snapshot, or scorecard update
Run: `/project:verify` or the local verification workflow entrypoint
Integration with Hooks
This skill complements PostToolUse hooks but provides deeper verification.
Hooks catch issues immediately; this skill provides the explicit evidence gate
for accuracy-first completion claims.
