# Tango pack for ECC

Tango is the server side of an agent harness: one task board, one identity per
agent, tamper-evident receipts, and a shared memory vault that every harness on
the team reads and writes. ECC gives your agent a disciplined local loop; Tango
gives the whole team a durable record of it.

## Install

Add the MCP server to whichever harness you run:

```bash
# Claude Code
claude mcp add --transport http tango https://tango.applayer.io/mcp

# Anything that reads an mcp-servers.json (Codex, Cursor, OpenCode, Kimi, Zed…)
curl -sL https://tango.applayer.io/ecc/mcp-servers.json -o ./mcp-servers.json
```

Then install the working skill so the agent follows the loop without being told:

```bash
mkdir -p ~/.claude/skills/tango-work-loop
curl -sL https://tango.applayer.io/ecc/tango-work-loop.md \
  -o ~/.claude/skills/tango-work-loop/SKILL.md
```

Sign in once through the MCP OAuth prompt. `whoami` then tells you which
organization, workspaces and agent identity you are operating as.

## How the ECC lifecycle maps onto Tango

| ECC step | Tango call |
| --- | --- |
| `/ecc:plan` produces a plan | `create_task` (with `subtasks`), or `request_decomposition` on an existing task |
| Plan approved | `claim_task` / `pull_next_task` — the lease is who is accountable |
| RED evidence captured | `add_artifact` with `evidence_kind: "tests"` |
| Implementation notes | `add_progress_note` |
| Fresh-context review | `handoff_task` to a reviewer, or `add_artifact` with `evidence_kind: "review"` |
| Verified and done | `complete_task` with `summary` + `evidence_artifact_ids` |
| Remember | `memory_save`, `log_project_decision`, or a lesson on the project |
| Recall in another harness | `memory_search` / `memory_read` |

## What Tango adds that a local harness cannot

- **Evidence gates.** A task declares the proof it must carry — plan, tests,
  screenshot, document, review. `complete_task` rejects a completion that is
  missing a required item, before a human ever looks.
- **Signed receipts.** Every completion is hash-chained and independently
  verifiable at `https://tango.applayer.io/verify`.
- **Learned instincts.** Send-backs and escalations are distilled into
  confidence-scored lessons that ride along in the next briefing.
- **A vault the team shares.** `memory_save` / `memory_search` / `memory_read`
  work across Claude, Codex, Cursor, Hermes and anything else on the team, with
  workspace scoping and access control instead of per-machine Markdown.
- **A configuration audit.** `security_posture` reports dormant keys,
  over-broad scopes, insecure webhooks and secrets pasted into task text.

MIT-licensed pack; Tango itself is a hosted service.