Updated: 2026-02-22 Version: 4.0 (Evidence Layer for the OpenCode Ecosystem)
Add the missing layer to AI agent tooling: structured evidence.
The OpenCode ecosystem has orchestration, policy, session management, and parallel execution. It has zero tools for proving what agents did, why, and whether the work was verified. SDP fills that gap.
A protocol and evidence layer, not a platform.
- The Protocol — prompts, JSON schemas, shell hooks that structure agent work into phases (Intent → Plan → Execute → Verify → Review → Publish). Language-agnostic. Works with OpenCode, Claude Code, Cursor.
- The Evidence Envelope — a strict 9-section JSON document (intent, plan, execution, verification, review, risk, boundary, provenance, trace) that every agent run produces. Validated by schema. Hash-chained for tamper detection.
- The PR Gate — one CLI command in CI that blocks merge unless evidence is complete and valid.
- Not an orchestrator (use Vibe Kanban, Swarm Tools)
- Not a policy engine (use Cupcake)
- Not a K8s operator (use kubeopencode — we contribute upstream)
- Not a session manager (use micode, oh-my-opencode)
SDP composes with all of these. It adds evidence to whatever workflow you already have.
- Individual developer — installs SDP as a submodule, gets structured agent workflow + evidence log
- Team using AI agents in CI — adds
sdp-evidence validateto PR gates, gets audit trail - Platform team on K8s — uses kubeopencode + SDP evidence bridge for agent-to-PR pipeline with proof
SDP is adoptable without "the whole spaceship":
| Level | What You Get | What You Install |
|---|---|---|
| Protocol only | Structured agent workflow via skills | curl install.sh — prompts, schemas, hooks |
| + Evidence | Audit log with hash-chain provenance | CLI: sdp log, sdp-evidence validate |
| + K8s Bridge | Agent runs on kubeopencode produce evidence | Adapter controller (in development) |
Each level is independently valuable. Upgrade path is additive.
┌─────────────────────────────────────────────────┐
│ OpenCode Ecosystem │
│ │
│ Orchestration: Vibe Kanban, Swarm Tools │
│ Policy: Cupcake │
│ K8s: kubeopencode │
│ Issues: Beads │
│ Sessions: micode, oh-my-opencode │
│ │
│ Evidence: SDP ◄── this is us │
│ │
└─────────────────────────────────────────────────┘
- Protocol spec with 12 skills (v0.9.8, 18 releases)
- Evidence log with hash-chain provenance
- CLI:
sdp doctor,sdp guard,sdp log,sdp status - Multi-agent review (6 agents), strategic planning (7 agents), OSS reality baseline analysis
- Install script with auto-detect for Claude Code, Cursor, OpenCode
- 1,004 commits, 16 stars, MIT license
| Priority | What | Status |
|---|---|---|
| P0 | Publish evidence JSON Schema in schema/ |
Ready |
| P0 | Release sdp-evidence CLI as standalone binary |
In progress |
| P1 | awesome-opencode listing | After first evidence CLI release |
| P1 | kubeopencode upstream PRs (retry budget, evidence hooks) | In progress |
| P2 | OpenCode plugin for local evidence collection | Planned |
| P2 | Adapter controller hardening for K8s bridge | In development |
Exploratory work, no promises:
- Multi-persona adversarial review with dissent tracking
- Agent self-improvement loops (failure pattern → adjustment)
- Cross-project federation via NATS
- Telemetry-driven backlog generation
Private repo sdp_lab — open an issue to request access.
| Metric | Current | Target |
|---|---|---|
| Protocol releases | 18 (v0.9.8) | v1.0 |
| Evidence CLI published | No | Yes |
| awesome-opencode listing | No | Yes |
| External users of evidence CLI | 0 | >= 1 |
| kubeopencode upstream PRs | In progress | >= 1 merged |
- Evidence is the moat. Everything else is ecosystem.
- Compose, don't replace. Use Vibe Kanban for orchestration, Cupcake for policy, kubeopencode for K8s.
- Ship what exists. Don't announce what isn't built.
- One developer, honest scope. Scale the community, not the claims.
- MIT forever. Public SDP layers stay open source.