Proposals

Improvement proposals from simulations

Accepted Proposals

ACCEPTED sim-001 | Iter #2 | 2026-03-11

Proposal #001: Standardized Deploy Manifest (deploy.yml)

Problem: Deploy pipeline has 0% automation. Sentinel collects deploy info ad-hoc via relay messages (8-12 msgs, 60-90 min per deploy). 28% of relay messages are duplicates.

Solution: Standardized deploy.yml manifest in each repository defining build config, containers, dependencies, domains, and required env vars.

Predicted Impact

KPICurrentPredictedChange
Deploy velocity60-90 min10-15 min-80%
Messages/deploy8-122-3-75%
Success rate~60%~85%+25pp
Automation0%~60%+60pp

Rollout Plan

PhaseTaskEffortStatus
1Define spec + create for s60-auth1 dayPending
2Sentinel manifest parser + validation2 daysPending
3Extend to pulse, s60-mail, billit2 daysPending
4Badwolf + Venom (special build)1 dayPending
5Auto-deploy on git pushFuturePlanned

Risks

ACCEPTED sim-002 | Iter #3 | 2026-03-12

Proposal #002: Security Hardening — Secret Management

Problem: 11 unique secret types (32+ occurrences) found in plaintext in relay message history. File permissions wrong on 55% of secrets files/directories. Zero secret rotation since initial setup.

Solution: 4-phase remediation: (1) Fix permissions immediately, (2) Rotate all exposed secrets, (3) Harden relay API with pattern detection & message expiration, (4) Establish 90-day rotation schedule.

Exposed Secret Types

Secret TypeRisk LevelFound In
ANTHROPIC_API_KEYCRITICALsentinel, infra
DO_API_TOKENCRITICALsentinel, infra
CF_API_TOKENCRITICALsentinel
GOOGLE_CLIENT_SECRETHIGHauth, sentinel
FACEBOOK_APP_SECRETHIGHauth, sentinel
JWT_SECRETHIGHauth, sentinel
MANAGED_PG_PASSWORDHIGHsentinel
REDIS_PASSWORDMEDsentinel
RESEND_API_KEYMEDmail, sentinel
JENKINS_DB_PASSWORDMEDsentinel
_AUTH_CLIENT_SECRETMEDauth

Predicted Impact

KPICurrentPredictedChange
Secrets in relay32+ plaintext0-100%
File permissions OK~55%100%+45pp
Secret rotationNever90 daysNew
Exposed secret types110-100%

Rollout Plan

PhaseTaskTimelineStatus
1Fix file permissions (secrets/ + servers)24hPending
2Rotate all 11 exposed secret types1 weekPending
3Relay API hardening (pattern detection, expiration)2 weeksPending
490-day rotation schedule + automated checksOngoingPlanned

Positive Findings

Side Findings (from deploy audit)

SeverityFindingAction
CRIT 6+ secrets in plaintext in relay history Proposal #002 (planned)
WARN Pulse /health returns 404 Sent to sentinel
WARN s60-mail has no nginx config on prod Sent to sentinel
WARN Relay API duplicates 28% of messages Bug report needed
ACCEPTED sim-029 | Iter #65 | 2026-03-15

Sim-029: Claude Code Feature Audit

Complete inventory of Claude Code v2.1.76: 39 CLI flags (8% used), 43 plugins (0 installed), 0 hooks, 2 broken MCP servers.

Phase 1 (SELF-IMPLEMENTED): run.sh flags: --no-session-persistence (saves 40MB), --name (session tracking), --max-budget-usd 0.75 (cost cap), --fallback-model sonnet (reliability).

Phase 2 (propose to Libor): Install hookify + security-guidance + code-review plugins. Remove broken Google MCP servers. --effort medium for token savings.

Phase 3 (ecosystem): --allowedTools per agent, TypeScript LSP for NestJS repos, custom S60 agents.

F-19566 stale sessions (40MB) — fixed by --no-session-persistence
F-1962 broken MCP servers (Google Calendar/Gmail)
F-1970/43 plugins installed
F-1980 hooks configured
F-199Feature utilization 8% (3/39)

Rejected (with learnings)

No rejected proposals yet.

Implemented (with KPI comparison)

No implemented proposals yet.