config-guardian

name

config-guardian

description

Universal governed config co-pilot. Before ANY change to ANY system: (1) check latest docs and running version (docs-first), (2) propose as diff with risk analysis, never apply directly (propose-only), (3) log every change with evidence and rollback (change ledger). Works for OpenClaw, Docker, PostgreSQL, Nginx, arifOS, or any software. Triggers on: 'change config', 'fix settings', 'update', 'propose patch', 'explain config', 'validate config', 'why did we change X'. Enforces propose-only workflow — human applies via git.

Config Guardian — Governed Config Co-Pilot

Rule #1: Read the manual. Rule #2: Propose diff, do not apply. Rule #3: Log everything.

One skill. Three modes. Universal.

---

Mode 1: Docs-First (Before Any Change)

Before touching ANY config, verify knowledge is current:

Step 1: Check running version

# Adapt to the system
openclaw status           # OpenClaw
docker --version          # Docker
psql --version            # PostgreSQL
arifos health             # arifOS

Step 2: Fetch latest docs

Use web_fetch or local docs (/app/docs/ for OpenClaw).

• Official docs site > README > changelog
• Check release notes between installed and latest version

Step 3: Compare

State explicitly:

• Running version: what's installed
• Docs version: what the fetched docs describe
• Delta: breaking changes, deprecations, new features

Situation	Action
Docs match knowledge	Proceed
Docs newer than knowledge	Proceed, cite fetched docs only
Can't fetch docs	HALT — tell human
Breaking changes found	HALT — show changes, ask human

---

Mode 2: Propose Config Change

Step 1: Read current config

cat ~/.openclaw/openclaw.json    # or whatever config file

Step 2: Generate unified diff

--- a/openclaw.json
+++ b/openclaw.json
@@ context @@
-  "dmPolicy": "pairing",
+  "dmPolicy": "allowlist",
+  "allowFrom": ["tg:267378578"],

Step 3: Include with every proposal

• What changes: plain language
• Before behavior: how it works now
• After behavior: how it will work after
• Risk: Low / Medium / High / Critical
• Rollback: exact steps to undo
• Docs referenced: URL or path that justified this change

Step 4: Human applies

For protected files, tell the sovereign:

# Apply on host via git
git apply /tmp/config-patch.diff
git diff
git commit -m "config: <description>"

Never use edit/write/apply_patch on Tier 1 files.

---

Mode 3: Validate Config

Check a proposed config or diff against docs:

• JSON5 syntax valid
• No secrets in plaintext (use ${VAR})
• Auth not "none" in production
• Model IDs reference valid providers
• Timezone correct
• Fallback chain has ≥2 entries

Output: VALID or INVALID with specifics.

---

Change Ledger (Built-In)

Every change that gets applied must be logged:

### Change: [description]
- **Date**: [ISO 8601]
- **System**: [software + version]
- **Docs referenced**: [URL/path]
- **Proposed diff**: [summary]
- **Risk**: [Low/Med/High/Critical]
- **Approved by**: [name]
- **Rollback**: [steps]

Store in: daily memory (memory/YYYY-MM-DD.md) + git commit message.

---

Protected Paths

Tier 1: Constitutional — NEVER modify directly

SOUL.md, USER.md, AGENTS.md, IDENTITY.md, core/, .env

Tier 2: Operational — Propose-only, sovereign applies

openclaw.json, opencode.json, CLAUDE.md, GEMINI.md, docker-compose.yml

Tier 3: Free — Agent can modify

memory/.md, logs/.jsonl, skills/*/SKILL.md

---

Risk Matrix

Risk	Examples
Low	Change model, add cron job
Medium	Change DM policy, add agent
High	Change auth, expose port
Critical	Rotate keys, delete data → 888_HOLD