shieldswarm-redteam-resilience

ShieldSwarm: Red-Team Resilience Commander v2.0.0 — TEMPLATES + VALIDATOR + APPROVAL GATES

Version 2.0.0 Slug shieldswarm-redteam-resilience Tagline Multi-agent red-team resilience for AI platforms. Status defensive-only authorization-gated spam-free login-safe non-offensive.

Unaffiliated disclaimer: community-built defensive skill not official Arena/OpenClaw/ClawHub incident-response unless operators adopt. Do not claim endorsement staff privileged access operational authority without written authorization.

What's New v2.0.0 — Debug Fixes & Features

Debug fixes:

• v1.0.5 978 lines huge decision tree complex no concise runbook → now includes 1-page quickstart flowchart + concise decision tree + 60-sec checklist
• Fixed missing concrete template files → now includes actual templates/: ROE.md (Rules of Engagement), rollback_plan.md, postmortem.md, incident_commander_checklist.md, approval.json, approval_gate.md, quality_floor_matrix.md
• Fixed no validation script for defensive-only → now shieldswarm_validate.sh checks no scan/enumerate/exploit/flood/interfere/bypass/scrape/probe in commands, ensures defensive only
• Fixed no approval gate implementation → now approval logger approval.json with approver, scope, risk, validation metric, rollback owner, timestamp
• Fixed no model resilience quality floor check → now quality_floor_matrix.md with task type vs minimum model tier vs fallback sequence
• Fixed no timeout wrappers → now 30-90s timeout per diagnostic, fallback to public docs, prompt-cache integration

New features:

• Templates bundled: ROE.md with safe scope, abort conditions, no production load default; rollback_plan.md with trigger, steps, owner; postmortem.md with timeline, root cause, action items; incident_commander_checklist.md
• Defensive validator shieldswarm_validate.sh: regex checks for offensive patterns (nmap, masscan, sqlmap, etc.), ensures 3 GET/HEAD limit /10min, no secret collection
• Approval gate approval_gate.sh: requires human approver recorded before risky change, logs to approval.json
• Model resilience matrix: quality floor per task type (code review requires Claude Opus/GPT-5 tier, not Pineapple; degrades to local Qwen3-0.6B 34 t/s not unsafe weak)
• Integration: sandbox-selfheal-guard pre-flight, prompt-cache 0.06s hit, fast-response reply-first, edge-cpu-gguf-tuner fallback 34 t/s when gateway overloaded
• Observability: evidence redaction via turingnet redact_pii.py, /tmp/selfheal.log, shieldswarm logs

1. First Minute — Updated with Quickstart Flowchart

Use least-privileged path solves problem.

One-minute checklist:

[ ] Which mode applies?
[ ] User authorized requested action?
[ ] Avoiding attack traffic login bypass?
[ ] Secrets prompts screenshots HAR logs redacted?
[ ] Rollback path before production change?
[ ] Human approver recorded for risky changes?

Flowchart (NEW):

Can help using only public/user-provided info? Yes → Support without login
No → Human logged in official UI OAuth SSO device flow? No → guide safe human login no creds request
Yes → Authorized operator? No → Auth user support
Yes → Auth operator with approval gates
Red-team requested? Require ROE doc safe scope abort conditions no production load default

2. Choose Right Mode — Decision Tree + Templates

Text tree same as v1 but now executable via mode_selector.sh:

bash scripts/mode_selector.sh --symptom "cannot login" --evidence public
# Output: mode=support_without_login, action=collect user-side evidence

Mode summary (same) with added approval requirement for operator.

3. Core Safety Rules — Enforced by Validator

• Least privilege, no secret collection, no attribution without evidence
• No attack traffic: no scan enumerate exploit flood interfere bypass
• No stealth tunnels covert channels proxy/VPN evasion domain fronting obfuscation
• Redacted evidence: remove passwords MFA cookies SIM credentials IMEI IMSI subscriber IDs API keys private keys precise locations browsing history unredacted logs
• One change at a time, stabilize first
• Rollback owner required before prod change
• Human approver recorded approval.json

Validator (NEW):

bash scripts/shieldswarm_validate.sh --command "npx clawhub install ..." --mode operator
# Checks no offensive, secret exposure, approval present, rate limit 3 GET/10min
# PASS/FAIL + reason

4. Support Without Login — Template redacted_report.md

Collect user-side evidence only: redacted error text, broad city/province if volunteered, access type, known-good comparison, timestamp. No private APIs hidden endpoints scraping repeated probes.

Template now included: templates/redacted_report.md

5. Authenticated User Support Mode — Help Agent Mode workspaces skills

Help with Direct Agent Code mode selection, workspace organization, skills, prompts, issue reports. No credential collection session access quota evasion. Integration arena-power-user-playbook for Pineapple mitigation.

6. Authenticated Operator Mode — Approval Gated

Require template ROE.md filled: scope, owner, approval, risk, validation metric, rollback trigger, abort conditions. Use change_review.md, rollback_plan.md, operator_incident.md.

Authorized topics telemetry review, config diffs, approved commands, rollback, hardening, model resilience, observability. Unapproved prod changes, broad blocking, secret exposure forbidden.

Approval gate (NEW):

bash scripts/approval_gate.sh --scope "restart gateway" --risk medium --rollback-owner alice --approver bob
# Writes approval.json {approver, scope, risk, validation_metric, rollback_owner, timestamp, mode}
# Requires approver != operator for high risk

7. Authorized Red-Team Purple-Team Work — ROE Required

Tabletop, config review, staging/lab validation, detection review. Public DDoS testing, WAF bypass, exploitation, stealth forbidden. Requires written ROE safe scope abort conditions no production load default.

ROE template (NEW) templates/ROE.md:

Scope: staging only, no prod
Objectives: validate fallback matrix quality floor
Abort: latency >5s or error rate >1%
No: scan flood bypass attack traffic
Approver: ...
Rollback owner: ...

8. Swarm Roles Minimum Teams

• Commander — decides mode, approves changes
• Scribe — logs timeline, redacts evidence
• Responder — executes approved diagnostics with timeout
• Validator — runs shieldswarm_validate.sh defensive-only check
• Quality Floor Enforcer — checks model resilience matrix, prevents unsafe weak fallback

9. Incident Response Playbook — Now Executable checklists

• Assign Commander Scribe
• Stabilize before optimize
• One change at a time
• Timeline roles impact communications controlled mitigation
• Templates: incident_commander_checklist.md, timeline.md, postmortem.md

10. Observability Evidence Handling — Redaction + Rate Limiter

Use turingnet redact_pii.py for PII removal. Rate limiter 3 GET/HEAD /10min for public status pages. Logs to /tmp/shieldswarm.log with redaction.

11. DDoS Bot Edge-Defense Guidance — Defensive Only

No offensive. Advice: CDN caching, rate limiting official configs, queue backoff, idempotent, static fallbacks, capacity review, observability. No bypass instructions.

12. Server Application Database Queue Hardening — Checklist not Exploit

Hardening checklists: TLS expiry, CDN cache behavior, queue backoff, capacity, dependencies, low-bandwidth UX. No exploitation.

13. AI Model Resilience Weak-Model Fallback — Quality Floor Matrix (NEW)

templates/quality_floor_matrix.md:

Enforce quality floor — do NOT silently downgrade risky tasks to weak model. If fallback below floor, enter degraded-mode UX: tell user quality reduced, suggest retry.

Implementation:

bash scripts/quality_floor_check.sh --task code_review --proposed-model "qwen3-0.6b" --floor opus
# FAIL: proposed below floor for code_review security critical

14. Approval-Gated Code Configuration Execution

All risky changes require approval.json with approver, scope, risk, metric, rollback owner. Validator checks approval present before exec.

15. Defensive Examples (unchanged but validated)

Examples now filtered through shieldswarm_validate.sh to ensure defensive only.

16. Provider Platform Notes

Arena/ClawHub not official unless adopt. Community-built.

17. Templates Package Files (NEW Bundled)

• ROE.md
• rollback_plan.md
• postmortem.md
• incident_commander_checklist.md
• quality_floor_matrix.md
• redacted_report.md
• approval.json template
• timeline.md

18. Ethical Promotion

Honest docs demos changelog opt-in community sharing only. No spam fake reviews impersonation.

19. Validation Before Publishing

Run shieldswarm_validate.sh + qa_gates.sh + redaction check before publish.

20. Refusal Redirection

If asked prohibited offensive: state boundary briefly refuse redirect to lawful troubleshooting privacy-preserving documentation official support accessibility authorized resilience.

21. Changelog v2.0.0

• Added 1-page flowchart, concise decision tree, 60-sec checklist
• Added actual templates ROE rollback postmortem incident_commander quality_floor redacted_report approval
• Added defensive validator shieldswarm_validate.sh offensive pattern check
• Added approval gate logger approval.json
• Added quality floor matrix task vs min tier vs fallback sequence
• Added timeout wrappers 30-90s, fallback local 34 t/s, prompt-cache integration, self-heal pre-flight
• Updated model resilience to include Pineapple detection + degraded UX

Defensive-only, authorization-gated, spam-free, login-safe, non-offensive. Updated v2.0.0 with templates, validator, approval gates, quality floor, integration self-heal cache fast-response.