ohad6k/viberaven/production-context

VibeRaven: Production Context

Repo context says what exists. Production context says what is dangerous.

Hard Rule

Keep memory small enough to reuse in the next agent action. Do not create a report dump.

File

Maintain:

.viberaven/production-context.md

Create it when allowed and missing. Otherwise propose the exact entry.

Loop

read -> isolate risk -> fix/propose -> verify -> record -> open action

1. Read the current file before production-sensitive work.
2. Identify one relevant boundary: release, auth, data/RLS, provider dashboard, deploy/env, job/webhook, billing, storage, or customer path.
3. Use repo evidence first: git status, tags, changelog, PR links, diffs, migrations, env examples, tests.
4. Use VibeRaven/provider MCP evidence when available for provider state, release history, and dashboard-adjacent receipts.
5. Record only what changes the next decision.
6. Separate repo fixes from provider/human actions.

When the output names Next skill:, continue with that VibeRaven skill unless user input, auth, or provider proof is required.

Use Next skill: go-live when context is recorded and the next step is deploy/live proof. Use Next skill: what-broke when context shows a version/release regression. Use Next skill: architecture-context when the risk reveals an unclear product boundary.

File Shape

# VibeRaven Production Context
 
## Current Release / Change Window
## Recent Changes
## Architecture Boundaries
## Provider Boundaries
## Migration And Data History
## Incidents And Rollback Notes
## Fragile Customer Paths
## Verification Receipts
## Open Provider Or Human Actions

Entry Shape

### YYYY-MM-DD - short label
 
-Change:
-Evidence:
-Boundary:
-Danger:
-Repo fix:
-Verification:
-Provider/MCP proof:
-Open action:

Use unknown from repo when provider state is not proven.

Output

Context read:
Context updated/proposed:
Boundary:
Repo action:
Provider/MCP action:
Next skill:
Next:

Mistakes

• Copying long logs instead of linking commands, files, PRs, screenshots, or receipts.
• Claiming provider dashboards are fixed by repo edits.
• Asking for passwords, tokens, cookies, private keys, signing secrets, or raw env values.
• Recording generic notes that will not change a future agent decision.