Files

T

sysadminandClaude Opus 4.8 1301a57de4 docs(governance): MCP restart governance and authorization policy (#656 )

Adds docs/architecture/mcp-restart-governance.md, the restart-governance/v1 ADR
defining who may restart the MCP control plane and under what conditions.

- Recovery ladder (reconnect -> rebind -> scoped restart -> full restart -> host)
  with restart stated as the last resort.
- Authorization matrix across author/reviewer/merger/reconciler/controller/
  operator/admin; no LLM worker role may perform or authorize a full or host
  restart.
- v1 authority decision recorded: controller approval + automated safety gates;
  quorum deferred to a superseding ADR.
- Break-glass path with pre-declared incident and mandatory post-hoc audit.
- Ambiguous policy state denies restart.
- Stable policy IDs RG-01..RG-08 for later enforcement code to bind to.

Cross-links the ADR from docs/safety-model.md and docs/webui-deployment.md, and
adds tests/test_mcp_restart_governance_docs.py asserting acceptance criteria 1-5.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

2026-07-23 19:16:28 -04:00

3.6 KiB

Raw Blame History

MCP Safety Model

This document outlines the safety requirements for all tools within the MCP monorepo.

1. Audit Logging and Confirmation

All mutating actions (e.g., triggering builds, creating resources, updating environments) must be recorded in an audit log. These actions require explicit confirmation from the user before execution to prevent accidental state changes.

2. Production Environment Safety

Any action that targets a production environment must have a hard confirmation gate. Production actions must never run based on vague or ambiguous prompts. The user must provide explicit, unambiguous consent to proceed with a production deployment or modification.

3. Secret Redaction

To maintain a secure environment, all secrets, tokens, passwords, and sensitive keys must be strictly redacted from:

System and application logs
Tool return values/outputs
Any form of persistent storage or console output

4. Read-Only First Policy

By default, MCP servers (such as jenkins-mcp and ops-mcp) operate in a read-only mode. Mutation capabilities are deny-by-default and fail-closed.

Note on naming: Historical design docs used jenkins-readonly / glitchtip-readonly skill names. Actual server packages are jenkins-mcp / glitchtip-mcp (registered via entry points in mcp-control-plane). See Gitea-Tools skills and mcp-control-plane #55 for registration.

5. Mutation Gating

Any mutating action (e.g., Gitea issue creation from GlitchTip, or Jenkins builds) must be explicitly allowed by the execution profile.

Jenkins build triggers are gated on a separate write boundary (jenkins-write-mcp / jenkins_mcp.write_server), not on the read-only jenkins-mcp surface. Triggers require a dedicated profile with jenkins.build.trigger, exact confirmation, and fail-closed mutation audit. No default profile carries trigger capability (#152 / mcp-control-plane #56).
GlitchTip to Gitea issue filing is a library-only orchestrator in mcp-control-plane (not on glitchtip-mcp). See #153 / mcp-control-plane #57.

6. Agent Commit Path (no improvised fallbacks)

When an author execution profile allows gitea.repo.commit and the gitea_commit_files tool is visible, agents must use that MCP path for repository commits. Fail closed instead of improvising alternate transports.

Forbidden when MCP commit is available:

WebFetch or other HTTP calls to external base64/decode services
Playwright or browser automation used to work around MCP commit
Manual LLM-generated base64 embedded in throwaway scripts as the primary commit transport

If shell helpers are unavailable and MCP commit cannot run, stop with a recovery report (restart session, clear hung terminals, use MCP-native commit). See llm-workflow-runbooks.md § MCP-native commit path (#260) and agent temp artifact cleanup (#261).

7. Process restart governance

Restarting the MCP control-plane process is destructive to concurrent multi-role work and is governed by a dedicated policy. Restart is a last resort behind narrower recoveries (reconnect, rebind), full/host restart is reserved to operator/admin under controller approval + automated safety gates, a unilateral LLM or operator full restart with active peers is forbidden, and ambiguous policy state denies restart. Break-glass is a separate, incident-backed path with a mandatory audit.

See architecture/mcp-restart-governance.md (#656) for the authorization matrix, the recovery ladder, break-glass conditions, and the RG-01–RG-08 policy IDs.

3.6 KiB Raw Blame History Unescape Escape