SOP — Operating Strix

docs / process / process-operate-strix

← all docs

SOP — Operating Strix

Owner: Security Engineering · Last Updated: 2026-08-14 · Review Cadence: quarterly, or on any major Strix version bump.

Purpose

Define how Strix is used as a recurring practice — not a one-off tool run — covering authorization, scan configuration, budget discipline, and findings handling.

Scope

Applies to any use of the open-source Strix CLI (self-hosted) or the managed app.strix.ai platform against systems the organization owns or has explicit written authorization to test. Does not cover incident response procedures for findings that turn out to indicate active compromise (see runbooks/review-and-triage-findings.md → Escalation).

RACI Matrix

Activity Responsible Accountable Consulted Informed
Requesting/confirming target authorization Security analyst Security lead Legal/compliance Engineering owner of target
Running scans Security analyst / CI pipeline Security lead — Engineering owner of target
Setting budget ceilings Security analyst Security lead Finance (for recurring/CI cost) —
Triaging findings Security analyst Security lead Engineering owner of target Leadership (for critical/high)
Remediation Engineering owner of target Engineering lead Security analyst Security lead
Model/provider credential rotation Whoever administers the toolchain Security lead — —

Process Flow

flowchart LR A["Confirm authorization\n& scope"] --> B["Configure scan\n(mode, budget, instructions)"] B --> C["Run scan\n(runbooks/run-a-detection-scan.md)"] C --> D["Triage findings\n(runbooks/review-and-triage-findings.md)"] D --> E["Route to remediation\nor CI gate (SARIF)"] E --> F["Re-scan to verify fix\n(fix-security-vulnerabilities-with-strix skill)"]

Detailed Steps

Step Who When How Output
Confirm authorization Security analyst Before any scan Written confirmation from the target's owner (email, ticket, or signed engagement doc) Authorization record
Choose scan mode Security analyst Per scan quick for CI/PR gating, standard/deep for full engagements — see strix/skills/scan_modes/ Scan config
Set budget ceiling Security analyst Per scan --max-budget / --max-budget-usd, sized to scan mode and target complexity Budget config
Run scan Security analyst / CI Per scan runbooks/run-a-detection-scan.md strix_runs/<run>/ artifacts
Triage Security analyst After every completed/stopped scan runbooks/review-and-triage-findings.md Prioritized findings list
Remediate Engineering owner Per finding, by severity SLA Standard engineering change process Fix committed
Verify Security analyst After fix ships Re-run Strix against the same target/scope Confirmed closure

Exceptions and Edge Cases

Case Handling
Scan hits budget/turn ceiling before finishing Not a failed scan — --resume continues it; check run.json's status/llm_usage.cost before treating an unfinished scan as "clean."
Target scope changes mid-scan Stop the running scan; do not rely on --instruction overrides to narrow scope mid-run — the system-verified scope block always takes precedence, but starting a fresh scan with corrected scope is cleaner.
CI PR-diff scans Use --scope-mode diff --diff-base origin/main (or rely on Strix's CI auto-scoping) so quick reviews only cover changed files.
Managed platform used instead of self-hosted Same triage/remediation process; findings arrive via app.strix.ai's dashboards/API instead of local strix_runs/.

Metrics

Metric Target
Time from scan completion to triage start < 1 business day for deep scans, < 1 hour for CI quick scans
% of findings with a working PoC (should be ~100% by design) 100% — a finding without one indicates a process problem, not just a false positive
Budget overrun rate (scans stopped by budget before finish_scan) Track and adjust default --max-budget per scan mode if trending up

STRIX — Process and Runbooks, 00_process-and-runbooks-index.md, runbooks/setup-and-credentials.md, runbooks/run-a-detection-scan.md, runbooks/review-and-triage-findings.md.