name: gsigmad-drift-scan description: "Scan all registered project adapters for CANON classification changes. Use when running scheduled drift detection or investigating a suspected CANON change. Loads adapters/*.md automatically and writes a drift report to .agent/drift_reports/." disable-model-invocation: true allowed-tools: [Read, Bash]

Drift Scan ($canon-drift-detection)

Scan all registered project adapters for CANON classification changes. Reads adapter metadata automatically from adapters/*.md (template at examples/adapters/example-project.md) — no manual project list needed. Writes a timestamped JSON drift report when classification changes are detected.

Usage

gsigmad-drift-scan

No parameters. All configuration is sourced from adapters/*.md automatically (per D-08, WIRE-03). Public release ships an example template at examples/adapters/example-project.md; copy to adapters/<your-project>.md to register a project for drift scanning.

Execution

Run the following Python block and display the result:

import sys
sys.path.insert(0, '.')
from gsigmad.governance.compiler.adapter_loader import load_project_registry
from gsigmad.governance.compiler.drift_scanner import scan_all_projects

# Load all registered project adapters from adapters/*.md (per D-08, WIRE-03)
project_registry = load_project_registry('.')

result = scan_all_projects(gsd_root='.', project_registry=project_registry)

print(f"Scan timestamp: {result['scan_timestamp']}")
print(f"Projects scanned: {result['projects_scanned']}")

if result['drift_detected']:
    print(f"\nDRIFT DETECTED -- {len(result['drift_events'])} event(s):")
    for event in result['drift_events']:
        print(f"  {event['project']}: {event['invariant_changed']} changed")
        print(f"    {event['old_classification']} -> {event['new_classification']}")
    print(f"\nFull report: {result['report_path']}")
else:
    print("No drift detected -- all CANON classifications stable.")

Output

• Summary to stdout: scan timestamp, projects scanned count, drift status.
• Full JSON report: written to .agent/drift_reports/DRIFT-{timestamp}.json only when drift is detected.
• Last scan marker: .agent/last_drift_scan.txt is always updated after every run (enables 24-hour cron trigger and threshold checks in drift_check.py).

Drift Report Schema

When drift is detected, the full report at .agent/drift_reports/DRIFT-{timestamp}.json follows this schema:

{
  "scan_timestamp": "2026-04-01T14:32:00Z",
  "projects_scanned": 8,
  "drift_detected": true,
  "drift_events": [
    {
      "project": "example-project",
      "invariant_changed": "classification",
      "old_classification": "EXPLORATORY",
      "new_classification": "CONFIRMATORY",
      "detected_timestamp": "2026-04-01T14:32:00Z"
    }
  ],
  "report_path": ".agent/drift_reports/DRIFT-20260401T143200Z.json"
}

Fields in each drift event:

• project: project name from the adapter registry
• invariant_changed: which CANON invariant classification changed
• old_classification: classification recorded in the previous snapshot
• new_classification: classification found in the current CANON.md scan
• detected_timestamp: ISO 8601 UTC timestamp when the drift was detected

Non-Negotiables

• Do NOT manually construct project_registry; always call load_project_registry() so that adapter file changes propagate automatically without code edits.
• Projects with has_canon=false are silently skipped by scan_all_projects() — this is correct behaviour, not an error. Projects without a CANON.md cannot experience classification drift.
• Do NOT suppress or modify drift events; display them verbatim from the result dict. The raw output is the audit record.
• This command is read-only except for writing the DRIFT JSON report and updating last_drift_scan.txt. It does not modify any CANON.md, experiment record, or KG document.