[DEFERRED] feat(mcp): agent-pipeline bridge — descoped per v0.4 kill criteria #3

Closed

varutasu wants to merge 2 commits from feat/agent-pipeline-bridge into main

varutasu commented

2026-06-11 18:23:17 -04:00

(Migrated from github.com)

Status: DEFERRED. Do not merge. Branch preserved as the resurrection point for Phase 2a if/when the conditions in agent-pipeline/docs/PHASE-2A-DESCOPE.md are met.

What's on this branch

feat/agent-pipeline-bridge carries the Echodo MCP bridge implementation from agent-pipeline v0.4 Phase 2a:

6 lifecycle MCP tools: create_convoy, create_brief, transition_convoy_status, log_convoy_event, query_manifest_status, reconcile_from_files
convoy_events Drizzle schema + generated migration 0010_wandering_the_professor.sql
L1/L3 agent-pipeline scaffolding (.cursor/rules/no-go-zones.mdc, .github/CODEOWNERS, .github/PULL_REQUEST_TEMPLATE.md, .github/workflows/agent-context-drift.yml, .github/workflows/pr-health-rollup.yml, scripts/log-convoy-event.sh, scripts/wt.sh)
.cursor/agents/echodo.config.json with workspace_slug=convoys-tasks
.agent-context-manifest.yml tracking 18 installed artifacts
9 L2 role files (.cursor/agents/role-*.md)
packages/database/package.json script fix to auto-load .env

Two commits: dab939b (bulk install) + 7c00b5b (L2 role files referenced by manifest).

Why it's deferred

Per agent-pipeline/docs/PHASE-2A-DESCOPE.md and analytics/v0.4-beta1-results.md §Observational data:

The designed 5-convoy experiment never ran as specified. What ran instead — 8 PRs merged on stwl-labs/tcg-vault Jun 5-11 (#126-#133), all file-only, no MCP bridge installed in any consumer repo — provided the empirical answer. The pipeline kept producing real, mergeable work without the bridge. §10 kill criterion #3 ("Echodo bridge worked end-to-end in ≥ 4 of 5 convoys") failed hard — the bridge wasn't installed anywhere during the experiment window.

Decision: descope Phase 2a; ship Phase 1a alone; restore + harden the metrics shim (separate PR on stwl-labs/tcg-vault#134). Resurrect Phase 2a only after one of the reversal conditions in PHASE-2A-DESCOPE.md becomes true.

Reversal conditions (when to merge this)

Per the descope doc:

Convoy work needs to be reviewable from a non-developer (design partner, PM, stakeholder). Echodo web UI is the answer.
Cross-convoy search becomes painful — grepping .convoys/*.md no longer scales.
convoy_events telemetry needs to power a dashboard (not just be data).
Multi-repo convoy aggregation becomes a real use case (single Echodo workspace projecting across tcg-vault + zest + tavernlight + colab).

None are true today. Likely candidates for v0.6+ when Phase 7 retros land and Phase 8 self-improvement loops emerge.

Where the decision lives

varutasu/agent-pipeline#1 — Phase 1a + descope rationale + analytics
varutasu/agent-pipeline:docs/PHASE-2A-DESCOPE.md — durable record (single source of truth)
varutasu/agent-pipeline:analytics/v0.4-beta1-results.md §Observational data — supporting evidence
This PR — searchable signal in the stwl-labs/ubiquitous-invention PR timeline

Made with Cursor

> **Status: DEFERRED. Do not merge.** Branch preserved as the resurrection point for Phase 2a if/when the conditions in `agent-pipeline/docs/PHASE-2A-DESCOPE.md` are met. ## What's on this branch `feat/agent-pipeline-bridge` carries the Echodo MCP bridge implementation from `agent-pipeline` v0.4 Phase 2a: - 6 lifecycle MCP tools: `create_convoy`, `create_brief`, `transition_convoy_status`, `log_convoy_event`, `query_manifest_status`, `reconcile_from_files` - `convoy_events` Drizzle schema + generated migration `0010_wandering_the_professor.sql` - L1/L3 agent-pipeline scaffolding (`.cursor/rules/no-go-zones.mdc`, `.github/CODEOWNERS`, `.github/PULL_REQUEST_TEMPLATE.md`, `.github/workflows/agent-context-drift.yml`, `.github/workflows/pr-health-rollup.yml`, `scripts/log-convoy-event.sh`, `scripts/wt.sh`) - `.cursor/agents/echodo.config.json` with `workspace_slug=convoys-tasks` - `.agent-context-manifest.yml` tracking 18 installed artifacts - 9 L2 role files (`.cursor/agents/role-*.md`) - `packages/database/package.json` script fix to auto-load `.env` Two commits: `dab939b` (bulk install) + `7c00b5b` (L2 role files referenced by manifest). ## Why it's deferred Per `agent-pipeline/docs/PHASE-2A-DESCOPE.md` and `analytics/v0.4-beta1-results.md` §Observational data: The designed 5-convoy experiment never ran as specified. What ran instead — 8 PRs merged on `stwl-labs/tcg-vault` Jun 5-11 (#126-#133), all file-only, no MCP bridge installed in any consumer repo — provided the empirical answer. The pipeline kept producing real, mergeable work without the bridge. §10 kill criterion #3 ("Echodo bridge worked end-to-end in ≥ 4 of 5 convoys") **failed hard** — the bridge wasn't installed anywhere during the experiment window. Decision: descope Phase 2a; ship Phase 1a alone; restore + harden the metrics shim (separate PR on `stwl-labs/tcg-vault#134`). Resurrect Phase 2a only after one of the reversal conditions in `PHASE-2A-DESCOPE.md` becomes true. ## Reversal conditions (when to merge this) Per the descope doc: 1. Convoy work needs to be reviewable from a non-developer (design partner, PM, stakeholder). Echodo web UI is the answer. 2. Cross-convoy search becomes painful — grepping `.convoys/*.md` no longer scales. 3. `convoy_events` telemetry needs to power a dashboard (not just be data). 4. Multi-repo convoy aggregation becomes a real use case (single Echodo workspace projecting across tcg-vault + zest + tavernlight + colab). None are true today. Likely candidates for v0.6+ when Phase 7 retros land and Phase 8 self-improvement loops emerge. ## Where the decision lives - `varutasu/agent-pipeline#1` — Phase 1a + descope rationale + analytics - `varutasu/agent-pipeline:docs/PHASE-2A-DESCOPE.md` — durable record (single source of truth) - `varutasu/agent-pipeline:analytics/v0.4-beta1-results.md` §Observational data — supporting evidence - This PR — searchable signal in the stwl-labs/ubiquitous-invention PR timeline Made with [Cursor](https://cursor.com)

varutasu commented

2026-06-11 18:23:18 -04:00

(Migrated from github.com)

Closing without merge per the body. Branch preserved for resurrection; do not delete. The descope decision is recorded in varutasu/agent-pipeline:docs/PHASE-2A-DESCOPE.md.

Closing without merge per the body. Branch preserved for resurrection; do not delete. The descope decision is recorded in `varutasu/agent-pipeline:docs/PHASE-2A-DESCOPE.md`.

github-actions[bot] commented

2026-06-11 18:23:27 -04:00

(Migrated from github.com)

Pipeline Health

CI gates

Gate	Status
CI: Lint, type-check, test	⏳ in progress

Build runs on Coolify post-merge; this CI runs lint + type-check + tests only. No per-PR preview URL by default.

Role reports

Role	Status
Reviewer report	⏳ pending
A11y audit	⏳ pending
Design system audit	⏳ pending

See individual comments above for details. This rollup updates automatically.

## Pipeline Health ### CI gates | Gate | Status | | --- | --- | | CI: Lint, type-check, test | ⏳ in progress | _Build runs on Coolify post-merge; this CI runs lint + type-check + tests only. No per-PR preview URL by default._ ### Role reports | Role | Status | | --- | --- | | Reviewer report | ⏳ pending | | A11y audit | ⏳ pending | | Design system audit | ⏳ pending | See individual comments above for details. This rollup updates automatically.