mirror of
https://github.com/affaan-m/ECC.git
synced 2026-09-18 07:37:59 +02:00
Affaan MustafaandClaude
b06c78cc6b
Packaged by Haley for the 2.1 launch. These must land on main before the v2.1.0 tag, because the announcement and README reference the assets by raw.githubusercontent.com URL on main. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0149VwNuynam6rvEfcMmiHHa
78 lines
2.5 KiB
Markdown
78 lines
2.5 KiB
Markdown
# ECC: The Agent Harness Operating System
|
|
|
|
> Plan artifact · generated by `/ecc:plan` · review in Plan Canvas, then approve to begin
|
|
|
|
## What ECC installs into your agent
|
|
|
|
Your agent can already write code. ECC gives it an engineering system: it plans
|
|
before it builds, verifies with tests, reviews its own work from a fresh context,
|
|
and turns repeated wins into reusable skills.
|
|
|
|
```mermaid
|
|
flowchart LR
|
|
P[Plan] --> T[Test first] --> I[Implement] --> R[Review] --> V[Verify]
|
|
V --> M[Remember] --> S[Improve skills] -.-> P
|
|
style P fill:#6885e8,color:#fff
|
|
style V fill:#6885e8,color:#fff
|
|
```
|
|
|
|
## How the pieces fit
|
|
|
|
| Concept | What it does | Context behavior |
|
|
|---|---|---|
|
|
| Skills | Reusable workflows: TDD, security review, deep research | Loaded when the task needs them |
|
|
| Agents | Scoped workers with their own context and tool permissions | Isolate planning, implementation, review |
|
|
| Rules | Durable project or language standards | Always loaded, so install selectively |
|
|
| Hooks | Scripts triggered by harness events | Run outside the model context |
|
|
| Instincts | Patterns learned from real sessions | Recalled when relevant |
|
|
|
|
```mermaid
|
|
flowchart TB
|
|
subgraph Harness["Any harness: Claude Code · Codex · Kimi · Cursor · OpenCode"]
|
|
A[Your agent]
|
|
end
|
|
subgraph ECC["ECC"]
|
|
SK[279 skills] --- AG[67 agents]
|
|
RU[Rules] --- HO[Hooks]
|
|
ME[Memory + instincts] --- AS[AgentShield]
|
|
end
|
|
A --> SK
|
|
A --> AG
|
|
```
|
|
|
|
## The review gate you are using right now
|
|
|
|
`/plan` ends with a hard confirmation gate. Plan Canvas moves that gate from a
|
|
wall of terminal markdown to this page. Point at the part you mean, annotate
|
|
it, and approve from here.
|
|
|
|
```mermaid
|
|
sequenceDiagram
|
|
participant U as You (browser)
|
|
participant C as Plan Canvas
|
|
participant A as Agent (terminal)
|
|
A->>C: open plan-canvas-demo.plan.md
|
|
U->>C: annotate "Split phase 2"
|
|
C->>A: feedback JSON
|
|
A->>C: revised plan (live-reload)
|
|
U->>C: Approve plan
|
|
C->>A: verdict: approve
|
|
A->>A: begin implementation
|
|
```
|
|
|
|
## Rollout tasks
|
|
|
|
- [x] Plan Canvas: annotate, chat, approve from the browser
|
|
- [x] Kimi Code install target with Moonshot AI
|
|
- [x] Verified self-host path on Itô GPUs
|
|
- [ ] Ship ECC 2.1 release + announcement
|
|
- [x] Demo video (you are watching it)
|
|
|
|
## Risks
|
|
|
|
| Risk | Mitigation |
|
|
|---|---|
|
|
| Plans reviewed as walls of text get skimmed | Canvas renders diagrams + anchored annotations |
|
|
| Same context writes and reviews code | Fresh-context reviewer agents |
|
|
| Harness config trusted by default | AgentShield scans the harness itself |
|