Connect installer to Stories 2/7 content (and move enrollment to a repo maintenance workflow in the .fullsend repo)

Makefile

@@ -100,7 +100,7 @@ e2e-test: e2e-playwright
 		$(MAKE) e2e-export-session; \
 		export E2E_GITHUB_SESSION_FILE="$(E2E_SESSION_FILE)"; \
 	fi; \
-	go test -tags e2e -v -count=1 -timeout 4m ./e2e/admin/
+	go test -tags e2e -v -count=1 -timeout 10m ./e2e/admin/
 
 e2e-export-session: e2e-playwright
 	@if [ -n "$$E2E_GITHUB_PASSWORD_FILE" ] && [ -z "$$E2E_GITHUB_PASSWORD" ]; then \

docs/ADRs/0007-per-role-github-apps.md

@@ -37,3 +37,4 @@ The installer checks for existing app installations before creating new ones. If
 - PEMs are write-once secrets: lost keys require app deletion and recreation.
 - The per-app model scales linearly with roles. Adding a new role means creating a new app — no shared credential rotation needed.
 - GitLab and Forgejo will need equivalent per-role identity mechanisms when their forge implementations are built.
+- **Permission escalation note:** The `fullsend` orchestrator app requires `workflows: write` (combined with `contents: write`) to create and update shim workflow files in enrolled repos during enrollment reconciliation. A compromise of this app's private key would allow an attacker to deploy arbitrary workflow files to all installed repos. Mitigations: (1) the app should be installed on specific repos, not org-wide; (2) the PEM is stored as a repo-level secret on `.fullsend`, limiting access to that repo's workflows; (3) per-role workflows validate `source_repo` against the config allowlist before generating tokens.

docs/architecture.md

@@ -39,8 +39,10 @@ Infrastructure platform choice and configuration are specified in the adopting o
 
 - Forge abstraction: all forge operations go through the `forge.Client` interface, keeping the rest of the codebase forge-agnostic ([ADR 0005](ADRs/0005-forge-abstraction-layer.md)).
 - Installation model: ordered layer stack (install forward, uninstall reverse, analyze for status reporting) with idempotent operations. Current stack: config-repo → workflows → secrets → inference → dispatch-token → enrollment ([ADR 0006](ADRs/0006-ordered-layer-model.md)).
-- Cross-repo dispatch: `workflow_dispatch` with an org-level dispatch token replaces `workflow_call`, keeping App PEM secrets in the config repo ([ADR 0008](ADRs/0008-workflow-dispatch-for-cross-repo-dispatch.md)).
+- Cross-repo dispatch: per-role `workflow_dispatch` workflows (`triage.yml`, `code.yml`, `review.yml`) with an org-level dispatch token, keeping App PEM secrets in the config repo ([ADR 0008](ADRs/0008-workflow-dispatch-for-cross-repo-dispatch.md)).
 - Shim workflow security: `pull_request_target` prevents PR authors from modifying the shim to exfiltrate the dispatch token ([ADR 0009](ADRs/0009-pull-request-target-in-shim-workflows.md)).
+- Repo maintenance: a workflow in `.fullsend` (`.github/workflows/repo-maintenance.yml`) reconciles enrollment shims in target repos when `config.yaml` changes or on manual dispatch. The CLI's `EnrollmentLayer.Install()` dispatches this workflow via `workflow_dispatch` and monitors it for completion, then reports any enrollment PRs created in target repos.
+- Installer scaffold: the `WorkflowsLayer` deploys content from an embedded scaffold (`internal/scaffold/`), keeping deployable files as real files under version control rather than Go string constants.
 
 **Open questions:**
 

docs/normative/admin-install/v1/adr-0012-fullsend-repo-files/SPEC.md

@@ -19,12 +19,24 @@ install of these layers:
 | Path | Layer responsible |
 |------|-------------------|
 | `config.yaml` | `ConfigRepoLayer` |
-| `.github/workflows/agent.yaml` | `WorkflowsLayer` |
-| `.github/workflows/repo-onboard.yaml` | `WorkflowsLayer` |
+| `.github/workflows/triage.yml` | `WorkflowsLayer` |
+| `.github/workflows/code.yml` | `WorkflowsLayer` |
+| `.github/workflows/review.yml` | `WorkflowsLayer` |
+| `.github/workflows/repo-maintenance.yml` | `WorkflowsLayer` |
+| `.github/actions/fullsend/action.yml` | `WorkflowsLayer` |
+| `.github/scripts/setup-agent-env.sh` | `WorkflowsLayer` |
+| `agents/triage.md` | `WorkflowsLayer` |
+| `env/gcp-vertex.env` | `WorkflowsLayer` |
+| `env/triage.env` | `WorkflowsLayer` |
+| `harness/triage.yaml` | `WorkflowsLayer` |
+| `policies/triage.yaml` | `WorkflowsLayer` |
+| `scripts/validate-triage.sh` | `WorkflowsLayer` |
+| `scripts/reconcile-repos.sh` | `WorkflowsLayer` |
+| `templates/shim-workflow.yaml` | `WorkflowsLayer` |
 | `CODEOWNERS` | `WorkflowsLayer` |
 
-Write order for workflow-related paths: `agent.yaml`, then `repo-onboard.yaml`,
-then `CODEOWNERS` (CODEOWNERS failure is non-fatal in the implementation).
+Write order: all scaffold files from `internal/scaffold/fullsend-repo/` in
+filesystem walk order, then `CODEOWNERS` last (CODEOWNERS failure is non-fatal).
 
 ## 3. Per-path requirements
 
@@ -35,19 +47,70 @@ then `CODEOWNERS` (CODEOWNERS failure is non-fatal in the implementation).
   (this SPEC only requires that the file exists as the tracked carrier for
   org configuration).
 
-### 3.2 `.github/workflows/agent.yaml`
+### 3.2 `.github/workflows/triage.yml`
 
-- **Contents:** exactly the document in
-  [`files/agent-dispatch-v1.yaml`](files/agent-dispatch-v1.yaml) (byte-for-byte
-  normative snapshot for v1).
+- Triage workflow triggered via `workflow_dispatch` with `event_type`,
+  `source_repo`, `event_payload` inputs.
 
-### 3.3 `.github/workflows/repo-onboard.yaml`
+### 3.3 `.github/workflows/code.yml`
 
-- **Contents:** exactly the document in
-  [`files/repo-onboard-v1.yaml`](files/repo-onboard-v1.yaml) (byte-for-byte
-  normative snapshot for v1).
+- Code workflow, same dispatch pattern as triage.
 
-### 3.4 `CODEOWNERS`
+### 3.4 `.github/workflows/review.yml`
+
+- Review workflow, same dispatch pattern as triage.
+
+### 3.5 `.github/workflows/repo-maintenance.yml`
+
+- Reconciles enrollment shims in target repos when `config.yaml` changes.
+
+### 3.6 `.github/actions/fullsend/action.yml`
+
+- Composite action: installs the fullsend CLI and runs the agent.
+
+### 3.7 `.github/scripts/setup-agent-env.sh`
+
+- Env prefix stripping helper sourced by per-role workflows.
+
+### 3.8 `agents/triage.md`
+
+- Agent instructions for the triage role.
+
+### 3.9 `env/gcp-vertex.env`
+
+- Environment variables for GCP Vertex AI inference.
+
+### 3.10 `env/triage.env`
+
+- Environment variables specific to the triage role.
+
+### 3.11 `harness/triage.yaml`
+
+- Harness configuration for the triage role.
+
+### 3.12 `policies/triage.yaml`
+
+- Policy definitions for the triage role.
+
+### 3.13 `scripts/validate-triage.sh`
+
+- Validation script for triage role output.
+
+### 3.14 `scripts/reconcile-repos.sh`
+
+- Shell script that reconciles enrollment shims in target repos. Called by
+  `repo-maintenance.yml`. Uses `gh` CLI and `yq` to read `config.yaml`.
+  For enabled repos: creates branches, writes shim workflows, and opens
+  enrollment PRs. For disabled repos: creates branches, deletes shim
+  workflows (via GitHub Contents API DELETE with blob SHA), and opens
+  removal PRs. Closes stale cross-direction PRs in both cases.
+
+### 3.15 `templates/shim-workflow.yaml`
+
+- The shim workflow template used by `scripts/reconcile-repos.sh` to write shim
+  workflows to target repos during enrollment and verify content during updates.
+
+### 3.16 `CODEOWNERS`
 
 - **Purpose:** grant the installing human ownership of all paths in the
   configuration repo.
@@ -66,9 +129,10 @@ includes `@` before the login as shown).
 ## 4. Relationship to implementation
 
 The reference implementation lives in `internal/layers/configrepo.go` and
-`internal/layers/workflows.go`. If implementation and this SPEC diverge, this
-SPEC is normative for **v1** tracked layout and file bodies (except
-`config.yaml` body per ADR 0011).
+`internal/layers/workflows.go`. The `WorkflowsLayer` deploys content from the
+embedded scaffold at `internal/scaffold/fullsend-repo/`. If implementation and
+this SPEC diverge, this SPEC is normative for **v1** tracked layout and file
+bodies (except `config.yaml` body per ADR 0011).
 
 ## 5. Out of scope (not Git-tracked in `.fullsend`)