diff --git a/openspec/changes/upstream-contribution/.openspec.yaml b/openspec/changes/upstream-contribution/.openspec.yaml
new file mode 100644
index 000000000..e14f3223d
--- /dev/null
+++ b/openspec/changes/upstream-contribution/.openspec.yaml
@@ -0,0 +1,2 @@
+schema: spec-driven
+created: 2026-04-13
diff --git a/openspec/changes/upstream-contribution/design.md b/openspec/changes/upstream-contribution/design.md
new file mode 100644
index 000000000..663930e2b
--- /dev/null
+++ b/openspec/changes/upstream-contribution/design.md
@@ -0,0 +1,155 @@
+# Design: Upstream Contribution Strategy
+
+## Architecture Decisions
+
+### AD-1: contextFiles type adaptation (v1.3.0)
+
+Upstream changed `ApplyInstructions.contextFiles` from `Record<string, string>` to `Record<string, string[]>` to support glob artifact outputs.
+
+**Impact on fork code:**
+- `src/core/pipeline/runner.ts` — reads contextFiles for gate synthesis context
+- `src/core/orchestration/resolver.ts` — reads contextFiles for task grouping
+- `src/commands/workflow/instructions.ts` — covers injection reads contextFiles
+
+**Resolution:** Handled during rebase (2026-04-13). All fork code now treats contextFiles values as `string[]`. The rebase was clean — auto-merge succeeded for `shared.ts` type change.
+
+### AD-2: update.ts transformer composition
+
+Fork adds `composeTransformers(overlayTransformer, toolTransformer)` pattern. Upstream added `pi` tool to the hyphen-command condition.
+
+**Resolution:** Merged both — `pi` added to condition, overlay composition preserved:
+```typescript
+const toolTransformer = (tool.value === 'opencode' || tool.value === 'pi')
+  ? transformToHyphenCommands : undefined;
+const transformer = composeTransformers(overlayTransformer, toolTransformer);
+```
+
+### AD-5: Workspace foundation impact (added 2026-05-04)
+
+Upstream merged `workspace-foundation` (PR #1045, archived as canonical spec) on 2026-05-04. This introduces a cross-repo planning model: workspaces live at `<global-data-dir>/workspaces/<name>/`, link multiple repos via stable names, and own `changes/` at the workspace root.
+
+Six slices are planned (1+2 landed, 3-6 proposed):
+
+| Slice | Status | Surface that conflicts with our PRs |
+|-------|--------|-------------------------------------|
+| `workspace-foundation` | landed | `instructions.ts`, `templates.ts`, `instruction-loader.ts`, `profiles.ts` — already merged into fork via sync/upstream-v1.3.1 |
+| `workspace-create-and-register-repos` | landed (CLI beta) | Setup CLI surface — no overlap with our PRs |
+| `workspace-open-agent-context` | proposed | Context injection — overlaps T1's plugin-loaded skill content |
+| `workspace-change-planning` | proposed | New change creation — neutral |
+| `workspace-apply-repo-slice` | proposed | **`apply` redefined as repo-slice implementation** — directly conflicts with T4's `openspec run start/complete` model |
+| `workspace-verify-and-archive` | proposed | **verify/archive at workspace + repo level** — directly conflicts with T2's phase extension model |
+
+**Strategic implications:**
+
+1. **T1 (Plugin Foundation)** — surface already churned by foundation slice. Rebase will be expensive but achievable. Submit early before slice 3 lands and adds another layer.
+
+2. **T2 (Workflow Phases)** — verify/archive phase types we want to add will likely be redefined by slice 6. Two paths:
+   - (a) Submit T2 anyway, position as "phase machinery that slice 6 can build on"
+   - (b) Wait for slice 6, then merge our verify/archive intent into the workspace model
+   - Default: (b) — saves rework, signals respect for upstream direction.
+
+3. **T4 (Pipeline Runner)** — `apply` semantics redefined by slice 5. `openspec run start/complete` would need to know whether it's running on a workspace change or a repo-local change. Defer until slice 5 stabilizes.
+
+4. **T3 (Orchestration Hints)** — workspace-neutral; flags affect agent dispatch regardless of change scope. Lower-risk; can ship after T1.
+
+5. **T5 (Fingerprinting)** — gate-checker internal, no workspace coupling. Independent.
+
+**CORE_WORKFLOWS union** (sync/upstream-v1.3.1 merge resolution): both `verify` (fork) and `sync` (upstream workspace) are now default-active. T2 PR upstream must reckon with this — either remove `verify` from CORE if upstream prefers minimal default, or argue for keeping both. This is a foreseeable point of friction.
+
+### AD-3: Odoo-specific content exclusion
+
+The following content exists in the fork but MUST NOT be included in any upstream PR:
+
+| Content | Location | Reason |
+|---------|----------|--------|
+| `schemas/odoo-sdd/` | Deleted from fork, lives in odoo-claude-code | Domain-specific |
+| `schemas/odoo-workflow/` | Deleted from fork, lives in odoo-claude-code | Domain-specific |
+| Odoo change proposals | `openspec/changes/` | Project-specific |
+| `openspec-parallel-merge-plan.md` | Root | Fork-specific planning doc |
+
+### AD-4: PR dependency chain
+
+```
+T0 (icebreaker) ── no code dependency
+  │
+  T1 (plugin foundation)
+  │  exports: PluginManifest, PluginLoader, HookDispatcher, ConfigValidator
+  │  new CLI: openspec plugin list|info
+  │
+  ├── T2 (workflow phases) ── depends on T1 schema types
+  │   extends: SchemaYaml with verify/archive phases
+  │
+  ├── T3 (orchestration) ── depends on T1 plugin loader
+  │   new CLI flags: --subagents, --teams, --sequential
+  │   new module: src/core/orchestration/
+  │
+  │   └── T4 (pipeline runner) ── depends on T1 + T3
+  │       new CLI: openspec run, openspec gate
+  │       new module: src/core/pipeline/
+  │
+  └── T5 (fingerprinting) ── weak dependency on T4 gate-checker
+      extends: gate-checker with content fingerprints
+```
+
+## Commit-to-PR Mapping
+
+### T1: Plugin Foundation
+| Fork Commit | Description |
+|-------------|-------------|
+| `a3a0270` | Plugin system core (loader, hooks, config, gates, schemas) |
+| `a037ec5` | Skill overlays injection |
+| `6ba2731` | README plugin documentation |
+| `354c7f2` | Archive add-skill-overlays change |
+| `fbec042` | Fix plugin-provided schema resolution |
+| `9446dbd` | Fix loadTemplate plugin schema resolution |
+
+### T2: Workflow Phase Extension
+| Fork Commit | Description |
+|-------------|-------------|
+| `11c22d2` | Configurable changesDir |
+| `63cae69` | Verify and archive phase schema support |
+
+### T3: Orchestration Hints
+| Fork Commit | Description |
+|-------------|-------------|
+| `34af9ac` | CLI-native orchestration hints |
+| `c552889` | Schema default_mode and --sequential flag |
+
+### T4: Pipeline Runner
+| Fork Commit | Description |
+|-------------|-------------|
+| `a324c3c` | Pipeline runner (runner, lock, types) |
+| `5e0c7db` | Fix hook command CWD resolution |
+| `df81ea0` | Covers auto-injection, change class routing |
+
+### T5: Fingerprinting
+| Fork Commit | Description |
+|-------------|-------------|
+| `375ce99` | Content-based fingerprinting for gate staleness |
+
+### Not included in any PR (Odoo-specific or fork-only)
+| Fork Commit | Description | Reason |
+|-------------|-------------|--------|
+| `acea753` | odoo-workflow schema | Odoo-specific |
+| `39f5636` | Lifecycle addon for odoo-workflow | Odoo-specific |
+| `c0f249b` | Align odoo-sdd schemas | Odoo-specific |
+| `1b0cc82` | Schema-level quality gates + odoo-sdd | Mixed (gates go in T1, odoo-sdd stays) |
+| `4c762cc` | 63 tests for gates | Goes with T1/T4 |
+| `9b3818e` | Test fixes for verify addition | Goes with T2 |
+| `13bb4e9` | ZshInstaller test isolation | Goes with T1 |
+| `22dde36` | Strengthen spec template format | Goes with T2 |
+| `778f1a5` | Ignore .actrc | Already in upstream via rebase |
+| `2c92b77` | Remove Odoo plugins from dev config | Fork cleanup |
+
+## Per-PR Checklist
+
+For each PR before submission:
+
+1. [ ] Create branch from latest upstream/main
+2. [ ] Cherry-pick relevant commits
+3. [ ] Remove any Odoo-specific content that leaked in
+4. [ ] Ensure all new exports are properly typed
+5. [ ] Run `npm run build` — clean compile
+6. [ ] Run `npx vitest run` — only pre-existing failures allowed
+7. [ ] Write PR description with context for upstream reviewers
+8. [ ] Cross-reference any relevant upstream issues (#879, #920, etc.)
diff --git a/openspec/changes/upstream-contribution/proposal.md b/openspec/changes/upstream-contribution/proposal.md
new file mode 100644
index 000000000..217cfded0
--- /dev/null
+++ b/openspec/changes/upstream-contribution/proposal.md
@@ -0,0 +1,64 @@
+# Upstream Contribution: Plugin System & Orchestration Platform
+
+## Problem
+
+OpenSpec is currently a static spec-driven development tool. It generates instruction files and manages change artifacts, but lacks:
+
+1. **Extensibility** — No way for teams to add custom lifecycle hooks, quality gates, or domain-specific schemas without forking
+2. **Quality assurance** — No automated validation that artifacts (proposal, specs, design, tasks) are aligned and traceable
+3. **Orchestration** — No mechanism to express parallel/sequential execution preferences for multi-agent workflows
+4. **Pipeline execution** — No runtime for automatically executing gates and hooks at lifecycle boundaries
+
+This fork has developed and production-tested these capabilities over 3 months of daily use in an Odoo 18 Enterprise development environment.
+
+## Proposed Solution
+
+Contribute the fork's 25 commits to upstream Fission-AI/OpenSpec as 5 sequential PRs, preceded by a Phase 0 icebreaker engagement. Each PR builds on the previous, and we wait for merge before opening the next.
+
+### PR Sequence (revised 2026-05-04 after workspace foundation landed)
+
+Upstream has shifted priority to a **cross-repo workspace model** (workspace-foundation merged in PR #1045 on 2026-05-04). Workspace slice 3–6 proposals overlap with our T2/T4 surface area. The original sequential plan is preserved below for context, but execution order and risk have been reassessed.
+
+| PR | Name | New order | Risk | Workspace overlap |
+|----|------|-----------|------|-------------------|
+| T0 | Icebreaker | DONE (PR #891 comment, TabishB ack 04-13, then 21+ days silent) | — | n/a |
+| **T1** | Plugin Foundation | **submit first** | **HIGH** (was MEDIUM) | Touches `init.ts/templates.ts/instructions.ts` — same surface workspace-foundation just refactored |
+| **T5** | Fingerprinting | **submit in parallel with T1** | LOW | None — gate-checker internal |
+| **T3** | Orchestration Hints | submit after T1 lands | LOW (was MEDIUM) | Cross-cutting, workspace-neutral |
+| T2 | Workflow Phase Extension (verify/archive) | **defer or reframe** | MEDIUM | **Direct overlap** with `workspace-verify-and-archive` slice 6 (proposed). Either reframe as "phase mechanism for slice 6" or fold in. |
+| T4 | Pipeline Runner | **defer until slice 5 lands** | HIGH | **Direct overlap** with `workspace-apply-repo-slice` slice 5 (proposed). `apply` semantics being redefined per repo-slice; T4's `openspec run start/complete` would need workspace-awareness. |
+
+**Why this order**: T1 + T5 + T3 ship the broad-utility primitives; T2/T4 are deferred until upstream commits to a workspace-friendly seam, otherwise we waste rebase effort fighting the wrong abstraction.
+
+### What stays internal (NOT contributed)
+
+- Odoo-specific schemas (odoo-sdd, odoo-bugfix, odoo-trivial, odoo-refactor)
+- alignment-check, dual-review, code-quality plugins (kept as independent examples)
+- Obsidian vault sync hooks
+- Git branch/worktree policy hooks
+
+## Success Criteria
+
+- All 5 PRs merged into Fission-AI/OpenSpec main
+- Upstream tests pass (excluding pre-existing `@inquirer/core` issue)
+- No Odoo-specific code in upstream
+- Plugin system documented in upstream README
+
+## Constraints
+
+- Conservative strategy: one PR at a time, wait for merge
+- Each PR must rebase onto latest upstream/main (which includes previous merged PR)
+- Upstream maintainer (TabishB) has ~24-48h review turnaround
+- Estimated total timeline: 3-6 weeks
+- `contextFiles` type changed from `string` to `string[]` in v1.3.0 — all fork code adapted during rebase
+
+## Risks
+
+| Risk | Mitigation |
+|------|-----------|
+| Upstream rejects plugin architecture | Phase 0 done (TabishB ack 04-13); silence since then is risk indicator — push T1 to force a review |
+| Workspace path eats T2/T4 surface | Defer T2/T4; reframe as workspace-aware after slice 5/6 stabilize |
+| API changes requested after PR1 | Each subsequent PR rebases from merged upstream, adapts as needed |
+| Long review cycles | Work on other projects during wait; conservative strategy reduces rework |
+| Workspace foundation blocks plugin auto-discovery for cross-repo | T1 must explicitly state "plugin-foundation is repo-local; workspace-level plugins are out of scope here" |
+| Maintainer attention saturated by workspace path | Submit T1 + T5 with explicit "complementary, not competing" framing; reference slice 1/2 architecture in PR body |
diff --git a/openspec/changes/upstream-contribution/tasks.md b/openspec/changes/upstream-contribution/tasks.md
new file mode 100644
index 000000000..be1cf23a8
--- /dev/null
+++ b/openspec/changes/upstream-contribution/tasks.md
@@ -0,0 +1,97 @@
+# Tasks: Upstream Contribution
+
+## 0. Icebreaker — Establish trust with upstream maintainer
+
+- [x] Review open P0 issues (#879, #881, #920) and identify one where plugin system insight is relevant
+- [x] Leave a thoughtful comment explaining how the fork's plugin architecture addresses the issue
+  - Posted on PR #891: https://github.com/Fission-AI/OpenSpec/pull/891#issuecomment-4234369497
+  - Connected to TabishB's "unify-template-generation-pipeline" vision
+- [x] Maintainer ack — 2026-04-13: TabishB replied "definitely be interested, let me take a look and get back to you"
+- [x] **Status (2026-05-04)**: 21 days silence. TabishB's attention shifted to workspace foundation (PR #1045 merged 05-04). Treat as **stalled, not negative** — push forward with T1 to force review.
+
+## 0.5. Sync upstream/main into fork (2026-05-04)
+
+- [x] Merge `upstream/main` into `sync/upstream-v1.3.1` branch
+  - 9 conflict files, 32 hunks; 4 real semantic decisions (CORE_WORKFLOWS union, instruction-loader nesting, etc.)
+- [x] All tests green: 1719/1719
+- [x] PR opened: https://github.com/stanleykao72/OpenSpec/pull/12
+- [x] Update this umbrella with workspace-aware revisions (this commit)
+
+## 1. PR: Plugin Foundation (T1) — submit FIRST
+
+Commits: a3a0270, a037ec5, 6ba2731, 354c7f2, fbec042, 9446dbd
+
+**Risk: HIGH** — surface area was just refactored by workspace-foundation (PR #1045). Rebase from latest upstream/main (after #1045 merge) and expect conflict resolution work in `init.ts/templates.ts/instructions.ts/instruction-loader.ts`. Reference sync/upstream-v1.3.1 PR #12 in fork as proof-of-concept that conflicts are tractable.
+
+- [ ] Create branch `feat/plugin-system` from latest upstream/main (post-#1045)
+- [ ] Cherry-pick and adapt commits (remove Odoo references)
+- [ ] Verify: `src/core/plugin/` (types, loader, hook-dispatcher, config-validator, context)
+- [ ] Verify: `src/commands/plugin.ts` (list, info subcommands)
+- [ ] Verify: skill overlay injection in `src/core/update.ts`
+- [ ] Verify: schema resolution supports plugin-bundled schemas
+- [ ] Include `schemas/spec-driven/schema.yaml` as default example
+- [ ] PR body: cite sync PR #12 conflict resolution; explicitly state "plugin-foundation is repo-local; workspace-level plugins are out of scope"
+- [ ] Run build + tests
+- [ ] `gh pr create -R Fission-AI/OpenSpec`
+- [ ] Record PR URL: ___
+- [ ] Track review feedback and iterate
+- [ ] Merged: [ ]
+
+## 2. PR: Fingerprinting & Staleness Detection (T5) — submit IN PARALLEL with T1
+
+Commits: 375ce99
+
+**Risk: LOW** — gate-checker internal, no workspace coupling. Submit in parallel with T1 (no dependency).
+
+- [ ] Create branch `feat/gate-fingerprinting` from latest upstream/main
+- [ ] Cherry-pick: content-based fingerprinting
+- [ ] Verify: SHA256 fingerprinting of tracked artifacts
+- [ ] Verify: stale gate detection in gate synthesis
+- [ ] Run build + tests
+- [ ] `gh pr create -R Fission-AI/OpenSpec`
+- [ ] Record PR URL: ___
+- [ ] Merged: [ ]
+
+## 3. PR: Orchestration Hints (T3) — submit AFTER T1 lands
+
+Commits: 34af9ac, c552889
+
+**Risk: LOW** (downgraded from MEDIUM) — workspace-neutral cross-cutting flags.
+
+- [ ] Rebase from upstream/main (now includes T1)
+- [ ] Create branch `feat/orchestration-hints`
+- [ ] Cherry-pick: CLI-native orchestration, default_mode, --sequential
+- [ ] Verify: `src/core/orchestration/` (types, resolver, domain-parser, group-builder)
+- [ ] Verify: --subagents, --teams, --sequential CLI flags
+- [ ] Run build + tests
+- [ ] `gh pr create -R Fission-AI/OpenSpec`
+- [ ] Record PR URL: ___
+- [ ] Merged: [ ]
+
+## 4. PR: Workflow Phase Extension (T2) — DEFERRED
+
+Commits: 11c22d2, 63cae69, 22dde36, 9b3818e
+
+**Risk: MEDIUM** — overlaps with `workspace-verify-and-archive` slice 6 (proposed). Default plan: wait for slice 6 to land, then refactor T2 into workspace-aware verify/archive support OR fold our intent into upstream's design.
+
+Decision gates before submitting:
+- [ ] Has slice 6 (workspace-verify-and-archive) landed in upstream?
+  - If yes → re-pitch T2 as a workspace-aware addition (or skip if slice 6 covers our use case)
+  - If no after 60+ days → submit T2 anyway as "phase mechanism for slice 6"; cite slice 6 proposal in PR body
+- [ ] CORE_WORKFLOWS friction: be ready to argue for keeping `verify` in CORE alongside upstream's `sync`
+
+(Sub-tasks unchanged when execution time arrives — see prior section content.)
+
+## 5. PR: Pipeline Runner + Gates/Hooks Execution (T4) — DEFERRED
+
+Commits: a324c3c, 5e0c7db, df81ea0, 1b0cc82 (gates portion), 4c762cc, 13bb4e9
+
+**Risk: HIGH** — `apply` semantics being redefined by slice 5 `workspace-apply-repo-slice` (proposed). `openspec run start/complete` must reckon with workspace vs repo-local change scope. Hard to ship before slice 5 lands.
+
+Decision gates before submitting:
+- [ ] Has slice 5 (workspace-apply-repo-slice) landed?
+  - If yes → refactor T4's runner to be workspace-aware; gate-checker reads from workspace `changes/` if applicable
+  - If no after 90+ days → submit T4 anyway as "pipeline mechanism for slice 5"; offer to add workspace-aware path in follow-up
+- [ ] Verify gate-checker remains decoupled from `apply` semantics (so slice 5 doesn't force gate-checker rewrite)
+
+(Sub-tasks unchanged when execution time arrives — see prior section content.)