Process Optimization: Scan → Remediate → Validate Workflow
Applies the standard current-state / waste / future-state / impact framework to the
process in process-operate-vvah.md.
1. Current State (as documented in the process doc)
Setup (once) → estimate → scan (S0–S9) → triage report.md (human) →
remediate (S10, edits code) → git diff review (human) → validate (S11) →
act on verdict (human) → merge/escalate
Eight steps, four of them requiring a human in the loop (triage, diff review, verdict action, escalation), three of them spending model budget (scan, remediate, validate).
2. Identify Waste
| Type | Where it shows up | Why it's waste |
|---|---|---|
| Waiting | Analyst triages report.md in a separate sitting from the scan; remediation often waits for a human to pick findings |
The scan → triage handoff has no forcing function — findings can sit unread for days between Steps 3 and 4 of the process doc. |
| Rework | remediate run without --mode report-only first, diff turns out unwanted, requires git checkout -- and a re-run |
Skipping the report-only preview (Step 1 alternative in the runbook) turns a review step into a revert-and-retry cycle. |
| Handoffs | Analyst runs scan → developer owns remediate/validate on their own repo → AppSec re-reviews needs_review verdicts |
Three handoffs per finding in the worst case; each is a queue with no SLA defined today. |
| Over-processing | Running --stop-after s9 scans repeatedly on an unchanged repo because --resume / caching isn't habitual |
Checkpointing exists (--resume, state DB) but is opt-in per invocation — easy to forget and re-pay the S1–S9 cost. |
| Manual work | Credential/profile setup (setup-and-credentials.md) is a manual, per-environment procedure; CI/CD wiring is entirely bespoke per team |
No shipped CI template — every team re-derives the same scan --stop-after s9 wrapper. |
| Cost drift | estimate is optional and easy to skip; actual token spend is only visible after the fact via util/metrics.py accounting, not surfaced mid-run |
Teams can commit to a full scan on a huge repo without a pre-flight cost gate. |
3. Future State (recommendations)
| Recommendation | Addresses | Effort |
|---|---|---|
Make estimate a required pre-scan step in the runbook and in CI (fail the job if projected tokens exceed a threshold) |
Cost drift | Low — process/tooling change only, no VVAH code change |
Default to --mode report-only for the first remediation pass per repo, require an explicit second command to apply |
Rework from unwanted diffs | Low — process convention, already supported by the CLI |
Add an SLA to the RACI: triage within 1 business day of scan completion; needs_review verdicts reviewed within 2 business days (already stated as a metric target in the process doc — formalize as a tracked SLA) |
Waiting, handoff drift | Low — ownership/process, no tooling change |
Ship a minimal CI job template (GitHub Actions/GitLab CI) wrapping scan --stop-after s9 + SARIF upload, as a reusable artifact in this project's samples/ |
Manual, duplicated CI wiring | Medium — one-time authoring effort, reusable across teams |
Habitual --resume: document it as the default invocation pattern (already true in run-a-detection-scan.md Step 4), and consider a wrapper script in samples/ that always adds --resume |
Over-processing / repeated cost | Low — script authoring, no VVAH change |
Batch setup across environments: script the .env + profile copy for repeat environments (dev/staging/prod runners) instead of manual setup-and-credentials.md walk-through each time |
Manual setup work | Medium — one scripted install path per environment class |
Track verdict distribution monthly (already a metric in the process doc) to catch a rising needs_review rate early — that's a leading indicator of a backend/config regression, not individual bad findings |
Handoff/rework caused by systemic issues going unnoticed | Low — reporting cadence, data already exists in DTOs |
4. Estimated Impact
- Cycle time (scan → verdict): formalizing the triage/review SLAs and defaulting to
--resumeshould cut the dominant source of delay (waiting between handoffs) without any change to VVAH itself — this is a process discipline win, not a tooling win. - Wasted spend: requiring
estimatebeforescan, andreport-onlybeforeremediate, removes the two most expensive rework paths (re-scanning unchanged repos, reverting unwanted fixes) at effectively zero implementation cost. - Consistency: a shared CI template removes N bespoke wrappers (one per team) converging
on the same
--stop-after s9pattern, reducing the chance a team accidentally runs a code-editing scan in CI.
5. Implementation Plan
- Update
process-operate-vvah.md's RACI/Metrics with the SLA additions above (process-only, do now). - Add a
--resume-by-default wrapper script and a CI job template undersamples/(next iteration). - Add a monthly verdict-distribution check to whoever owns the Metrics section's cadence.
- Revisit this document each time the process-doc's Review Cadence triggers (quarterly).