# Implementation Plan 0007: Deploy D&D Assets and Integrate Scriptorium

Status: ready

Target executor: GPT-5.6-Terra

Feature roadmap:
[Feature 0007: D&D Asset Deployment and Scriptorium Integration](features/0007-dnd-asset-deployment-and-scriptorium-integration.md)

Active plan location: `docs/roadmap/implementation.md`

## Objective

Implement the asset layer selected by Feature 0007. Correct the known-invalid
campaign and prompt references in their authoritative repositories, extend
`rakestraw.dnd_pipeline` with the six approved asset inputs and three direct
desired-state tasks, connect the deployed roots to Narratio and Scriptorium,
and prove the assembled behavior through native application interfaces and an
immediate clean repeat.

The feature roadmap remains authoritative for scope, rationale, target end
state, public variables, filesystem layout, and acceptance criteria.

## Execution Rules

- Execute Stage 1 through Stage 4 in numeric order.
- Immediately before each stage, read every document in that stage's Required
  Policy Review, then confirm its preconditions and the owning repository's
  working-tree state. Do not rely on policy recalled from an earlier stage.
- Modify only the repository named for the active stage. Preserve unrelated
  changes and stop if that repository is unexpectedly dirty.
- Correct asset defects in their source repositories; do not compensate for
  them with pipeline-role copies, rewrites, symlinks, or compatibility trees.
- Implement the production role with one `ansible.builtin.file` task and two
  `ansible.builtin.git` tasks. Do not add preliminary probes, registered-result
  conditionals, handlers, revision queries, checksums, manifests, cleanup
  commands, or recursive metadata passes.
- Use task-level `become: true` for the three privileged pipeline tasks. Do not
  require play-level escalation.
- Configure the existing application roles only through their public
  variables. Do not modify or reproduce their templates or tasks.
- Do not set `scriptorium_profile_dir`, invent execution profiles, select
  model backends, enable Notarius, or add the later opinionated D&D processing
  policy.
- Let `ansible.builtin.git` own creation, update, tracked-content convergence,
  and change reporting. Use the roadmap-selected `update: true` and
  `force: true`; do not separately purge untracked files.
- Keep tests at the suite-content boundary. Use explicit local Git origins and
  revisions for controlled inputs, then observe campaign and prompt discovery
  through Narratio and Scriptorium. Do not inspect Git internals or re-test the
  file and Git modules.
- Retain the existing Feature 0006 composition and explicit-version coverage.
  Do not reproduce any standalone role's installation or configuration
  matrix.
- Never place credentials or secret values in content, plans, examples, or
  tests.
- Run every stage's validation before committing it. Do not begin a dependent
  stage after an earlier stage fails.
- Commit each completed stage with its specified plain-English message. Do not
  push during out-of-band implementation.

## Starting State and Preconditions

- Feature 0006 is complete and independently reviewed.
- This plan, the agreed Feature 0007 roadmap, the corresponding MVP decisions,
  and workspace status are committed before out-of-band execution begins.
- `assets/narratio-prompts` is clean on `main` at
  `eef72b7d822a97d3e398a05f48bbfa20a2a6c109`, tracking `origin/main`
  without divergence.
- `assets/campaigns` is clean on `main` at
  `bb0aedf36a9a510d90a36e4df6b2c5c63135c025`, tracking `origin/main`
  without divergence.
- `ansible-role-dnd-pipeline` is clean on `main` at
  `4ea43580f66867d95ac23804554b07b67dbbaaee`, tracking `origin/main`
  without divergence.
- The local controller and managed-host proxy are Ubuntu 26.04 `x86_64`, with
  gathered facts and passwordless task-level privilege escalation available.
  Local execution remains the approved proxy for both supported host
  architectures during this MVP.
- Workspace tooling is available in `.venv`: Ansible Core 2.21.3 and
  ansible-lint 26.8.0. Scriptorium `v0.13.0` and Narratio `v2.0.0` are
  available through the completed roles for native validation.
- `community.general` 13.3.0 and Go Snap channel `1.26/stable` remain
  available. The existing pipeline integration entry point passes before
  implementation begins.
- The two local asset repositories are valid Git origins. Stage 4 may use
  their absolute `file://` URLs and explicit commit revisions without copying
  their content into the pipeline-role repository.
- The prompt repository contains no execution profiles. The current source
  defects are limited to the four references identified by the feature
  roadmap:
  - two uses of `../common/transcript.previous_recap.md` must name the existing
    `../common/previous_recap.user.md` file;
  - `pipeline-dnd/analyze/scene_map.schema.json` must resolve as
    `analyze/scene_map.schema.json`; and
  - `automation-pipeline-dnd/trim/bounds.schema.json` must resolve as
    `trim/bounds.schema.json`.
- Both campaign definitions contain an obsolete `inputs.players_file` field
  that Narratio 2.0 rejects. Their existing `inputs.party_file` fields remain
  the canonical party sources, but the referenced documents still use the
  legacy `party` list rather than `narratio.party.v1`.

## Repository Scope

Repositories that change:

- `assets/campaigns`: owns the two campaign-input corrections and two party
  schema migrations in Stage 1.
- `assets/narratio-prompts`: owns the four prompt-definition path
  corrections in Stage 2.
- `ansible-role-dnd-pipeline`: owns the public asset interface, downstream
  role wiring, direct deployment tasks, README, and integration coverage in
  Stages 3 and 4.

Repositories that may be inspected but must remain unchanged:

- `ansible-workspace-dnd`: authoritative roadmap, plan, policies, inventory,
  and status during out-of-band execution.
- `ansible-role-narratio` and `ansible-role-scriptorium`: authoritative public
  role interfaces.
- The other standalone application roles and all five application source
  repositories.
- `ansible-collection-dnd`: future collection assembly target.
- `rakestraw-ansible`: historical style and deployment reference.
- Every other repository in `workspace.yml`.

## Stage 1: Align Campaign Definitions with Narratio 2.0

### Objective

Make both authoritative campaigns valid under Narratio 2.0's canonical
party-input contract by removing the obsolete input and mechanically migrating
the party documents without deleting legacy source files.

### Required Policy Review

Before beginning this stage, read:

- `AGENTS.md`
- `docs/orientation.md`
- `docs/conventions.md`
- `docs/testing.md`

### Preconditions

- `assets/campaigns` is at the starting revision above and has no uncommitted
  changes.
- Both `centaurs/campaign.yml` and `dilfs/campaign.yml` contain a valid
  `inputs.party_file` value and the obsolete `inputs.players_file` value.
- Both referenced `party.yml` files contain the legacy list entries documented
  by the feature roadmap and have not already been partially migrated.
- `/usr/local/bin/narratio` is available for native campaign validation. If it
  is unavailable, stop and report the missing completed-role prerequisite
  rather than modifying application source or installing an unplanned
  substitute.

### Owning Repository

`assets/campaigns`

### Required Changes

1. Remove only the `inputs.players_file` entry from `centaurs/campaign.yml`
   and `dilfs/campaign.yml`. Retain each `party_file` entry and all other
   campaign fields unchanged.
2. Replace each legacy `party` list with a `narratio.party.v1` document using
   `schema_version` and a `characters` map.
3. Use exactly these stable character IDs: `krek`, `rick_oathbreaker`,
   `tlann`, `torvass_oathbreaker`, `hrank_greenfield`, `alan_hohey`,
   `svend_balderk`, and `popov`.
4. For every character, preserve the authored player name and character
   display name. Convert the single legacy class and level into one entry in
   `character.classes`. Correct Popov's evident class typo from `pladin` to
   `paladin`.
5. Omit the legacy `category: pc` field because character membership is
   implicit in the native map. Preserve each free-form legacy note as an
   adjacent YAML comment because the strict native schema has no note field.
6. Do not delete or rewrite either `players.yml` file. Removing or migrating
   unused legacy files is not required for native campaign validity and is
   outside this feature.

### Validation

Create `/tmp/dnd-pipeline-feature-0007.yml` with this temporary minimal
configuration:

```yaml
workspace:
  root: /tmp/dnd-pipeline-feature-0007-workspace
campaigns:
  root: /home/eric/Workspace/ansible-dnd/assets/campaigns
spool:
  root: /tmp/dnd-pipeline-feature-0007-spool
cache:
  root: /tmp/dnd-pipeline-feature-0007-cache
whisperx:
  transcribe_url: http://127.0.0.1:8000/transcribe
```

From `assets/campaigns`, run:

```text
/usr/local/bin/narratio config validate --config /tmp/dnd-pipeline-feature-0007.yml --campaign centaurs
/usr/local/bin/narratio config validate --config /tmp/dnd-pipeline-feature-0007.yml --campaign dilfs
git diff --check
git status --short
```

Delete only the temporary configuration after validation. Expected result:
Narratio natively validates both campaigns by registry ID; whitespace
validation passes; and only the two `campaign.yml` and two `party.yml` files
are modified.

### Completion Criteria

- Neither campaign definition contains `inputs.players_file`.
- Both retain `inputs.party_file`, whose document uses
  `schema_version: narratio.party.v1` and the roadmap's fixed character IDs.
- Player, character, class, level, and note information survives the migration
  in supported native fields or adjacent comments, with only the selected
  `paladin` typo correction.
- Both campaigns pass Narratio's native configuration validation.
- No `players.yml` or unrelated asset changes.

### Commit

Commit intent: `Align campaign definitions with Narratio`

## Stage 2: Correct the Prompt Library's Internal References

### Objective

Make the authoritative prompt repository internally consistent with its own
`prompts/` and `schemas/` layout so the pipeline role can deploy it without
content transformation or compatibility artifacts.

### Required Policy Review

Before beginning this stage, read:

- `AGENTS.md`
- `docs/orientation.md`
- `docs/conventions.md`
- `docs/testing.md`

### Preconditions

- Stage 1 is committed and `assets/campaigns` is otherwise clean.
- `assets/narratio-prompts` is at the starting revision above and has no
  uncommitted changes.
- The existing shared content file is
  `prompts/analyze/common/previous_recap.user.md`.
- The existing schema files are `schemas/analyze/scene_map.schema.json` and
  `schemas/trim/bounds.schema.json`.
- `/usr/local/bin/scriptorium` is available for read-only native prompt
  inspection. If it is unavailable, stop and report the missing completed-role
  prerequisite rather than modifying application source or installing an
  unplanned substitute.

### Owning Repository

`assets/narratio-prompts`

### Required Changes

1. In
   `prompts/analyze/character_meta_analysis/character_meta_analysis.yml` and
   `prompts/analyze/session_recap/recap.yml`, replace the nonexistent shared
   content reference with `../common/previous_recap.user.md`.
2. In `prompts/analyze/scene_map/scene_map.yml`, set `output.schema_path` to
   `analyze/scene_map.schema.json`, relative to the configured schema root.
3. In `prompts/trim/bounds.yml`, set `output.schema_path` to
   `trim/bounds.schema.json`, relative to the configured schema root.
4. Make no other editorial, prompt-content, version, profile, schema-content,
   or layout changes. Do not rename files or add compatibility links.

### Validation

From `assets/narratio-prompts`, run:

```text
/usr/local/bin/scriptorium inspect prompt --prompt-dir "$PWD/prompts" --prompt dnd.character_meta_analysis --prompt-version 2.0.0 --format json
/usr/local/bin/scriptorium inspect prompt --prompt-dir "$PWD/prompts" --prompt dnd.session_recap --prompt-version 2.0.0 --format json
../../.venv/bin/python - <<'PY'
from pathlib import Path

import yaml

root = Path.cwd()
definitions = (
    root / "prompts/analyze/scene_map/scene_map.yml",
    root / "prompts/trim/bounds.yml",
)
for definition in definitions:
    document = yaml.safe_load(definition.read_text())
    schema = root / "schemas" / document["output"]["schema_path"]
    if not schema.is_file():
        raise SystemExit(f"missing schema for {definition}: {schema}")
PY
git diff --check
git status --short
```

Expected result: Scriptorium loads both definitions that previously named the
missing shared file; both schema paths resolve to real files beneath
`schemas/`; whitespace validation passes; and only the four intended YAML
definitions are modified.

### Completion Criteria

- All four stale paths use the repository's actual shared-content and schema
  layout.
- Both corrected shared-content definitions load through Scriptorium's native
  inspection interface.
- Both schema references resolve relative to the repository's `schemas/`
  directory.
- No prompt semantics, IDs, versions, profiles, schemas, or unrelated assets
  change.

### Commit

Commit intent: `Correct prompt asset references`

## Stage 3: Add Minimal Asset Deployment to the Pipeline Role

### Objective

Implement the complete production asset interface, application-role wiring,
and module-driven deployment flow without adding test or verification logic to
role convergence.

### Required Policy Review

Before beginning this stage, read:

- `AGENTS.md`
- `docs/orientation.md`
- `docs/conventions.md`
- `docs/ansible.md`

### Preconditions

- Stage 1 and Stage 2 are committed, and both asset repositories are otherwise
  clean.
- Record both asset commits for Stage 4's explicit local revisions.
- `ansible-role-dnd-pipeline` is at the starting revision above and has no
  uncommitted changes.
- The pipeline role's existing Feature 0006 defaults, argument
  specifications, dependency declarations, and empty task file remain intact.
- The completed Narratio role accepts `narratio_campaigns_root`; the completed
  Scriptorium role accepts `scriptorium_prompt_dir` and
  `scriptorium_schema_dir`.

### Owning Repository

`ansible-role-dnd-pipeline`

### Required Changes

1. Add exactly these six defaults to `defaults/main.yml`:
   - `dnd_pipeline_campaigns_repository_url` with
     `https://gitea.maximumdirect.net/eric/campaigns.git`;
   - `dnd_pipeline_campaigns_revision` with `main`;
   - `dnd_pipeline_campaigns_root` with
     `/usr/local/share/narratio/campaigns`;
   - `dnd_pipeline_prompt_library_repository_url` with
     `https://gitea.maximumdirect.net/eric/narratio-prompts.git`;
   - `dnd_pipeline_prompt_library_revision` with `main`; and
   - `dnd_pipeline_prompt_library_root` with
     `/usr/local/share/narratio/prompt-library`.
2. Add matching argument-specification entries with concise descriptions.
   Use `str` for origins and revisions and `path` for roots. All six are
   required at role invocation because defaults always define them.
3. In `meta/main.yml`, pass `dnd_pipeline_campaigns_root` to the Narratio
   dependency as `narratio_campaigns_root`.
4. In the Scriptorium dependency, pass the prompt root's `prompts` child as
   `scriptorium_prompt_dir` and its `schemas` child as
   `scriptorium_schema_dir`. Use clear Jinja path expressions and do not set
   `scriptorium_profile_dir`.
5. Replace the empty production task list with exactly three tasks:
   - declare `dnd_pipeline_prompt_library_root` as a directory owned by
     `root:narratio` with mode `0755` using `ansible.builtin.file`;
   - converge the campaign Git origin and revision directly at
     `dnd_pipeline_campaigns_root`; and
   - converge the prompt-library Git origin and revision directly at
     `dnd_pipeline_prompt_library_root`.
6. For both `ansible.builtin.git` tasks, set `repo`, `dest`, and `version` from
   the corresponding public variables and set `update: true` and
   `force: true`. Apply `become: true` to all three tasks.
7. Do not duplicate the campaign-root directory declaration already owned by
   the Narratio dependency. Do not register task results, add `when`
   conditions, run handlers or commands, normalize repository contents
   recursively, purge untracked files, or validate content during production
   convergence.
8. Update the README to document automatic campaign and prompt-library
   deployment, all six asset variables and defaults, branch-following and
   immutable-commit selection, the direct installed layouts, Narratio and
   Scriptorium path wiring, and the remaining profile/backend/credential and
   D&D-policy deferrals.
9. Clarify that callers disabling common dependency management must provide
   both Git and Go. Remove Feature 0006 language claiming campaign and prompt
   content are deferred, but do not document unimplemented operators, secrets,
   removal, profiles, or end-to-end processing.

### Validation

From `ansible-role-dnd-pipeline`, run:

```text
../.venv/bin/ansible-playbook -i tests/inventory --syntax-check tests/converge.yml
../.venv/bin/ansible-lint --offline --profile production tests/converge.yml
git diff --check
git status --short
```

Expected result: Ansible accepts the expanded public interface, dependency
wiring, and three-task production role; production-profile lint passes; and
only intended defaults, metadata, task, and README files are modified. Full
asset convergence is deferred to Stage 4's controlled integration inputs.

### Completion Criteria

- Defaults and argument specifications expose exactly the six selected asset
  inputs in addition to the existing Feature 0006 interface.
- Narratio receives the same campaign root the Git task populates.
- Scriptorium receives the deployed `prompts/` and `schemas/` children and no
  profile directory.
- Production convergence consists of one direct directory declaration and two
  direct Git declarations, all with task-level privilege escalation.
- README documentation matches the implemented asset contract and remaining
  deferrals.
- Syntax, lint, and whitespace checks pass without test-specific production
  logic.

### Commit

Commit intent: `Deploy D&D pipeline assets`

## Stage 4: Add Focused Asset Integration Coverage

### Objective

Demonstrate that the real pipeline role deploys controlled campaign and prompt
repositories, connects them to the native applications, and coalesces cleanly
without testing trusted module internals or freezing mutable defaults.

### Required Policy Review

Before beginning this stage, read:

- `AGENTS.md`
- `docs/orientation.md`
- `docs/conventions.md`
- `docs/ansible.md`
- `docs/testing.md`

### Preconditions

- Stage 1 through Stage 3 are committed, and all three owning repositories are
  otherwise clean.
- Stage 3 static validation passes.
- The Stage 1 campaign commit and Stage 2 prompt commit are available from the
  local Git repositories.
- The existing integration driver still bootstraps exact application-role and
  collection dependencies into private temporary paths, exercises the
  Feature 0006 baseline and explicit Seriatim selection, and restores the
  baseline suite during cleanup.
- No test requires network access to the two default content origins. Network
  access needed by the existing application-role dependency and source-build
  flow remains unchanged.

### Owning Repository

`ansible-role-dnd-pipeline`

### Required Changes

1. Extend the existing `tests/integration` entry point rather than adding a
   new framework or scenario tree.
2. Resolve the absolute local paths of `../assets/campaigns` and
   `../assets/narratio-prompts` relative to the pipeline repository. Resolve
   each repository's current commit with read-only Git commands.
3. Add explicit baseline extra variables selecting those two local Git origins
   with `file://` URLs and their resolved immutable commits. Reuse the same
   explicit asset inputs for every convergence and cleanup invocation.
   Continue to supply the test WhisperX URL.
4. Do not assert that either selected commit or `main` is the role's current
   default. The explicit revisions are controlled test inputs, not snapshots
   of suite policy.
5. After baseline convergence, use Narratio's native configuration validator
   as the `narratio` account with the managed pipeline configuration and a
   campaign ID present in the deployed campaign registry, such as `centaurs`.
   Replace the less specific campaign-free validation where redundant.
6. Use Scriptorium's native `inspect prompt` interface with its managed
   configuration to load a deployed prompt whose repaired shared content is
   required, such as `dnd.session_recap` version `2.0.0`.
7. Inspect one deployed JSON-schema prompt through Scriptorium and verify that
   the schema path reported by that native definition resolves beneath the
   configured deployed schema directory. This is an integration assertion
   about the prompt-library contract; do not inspect Git metadata or duplicate
   the Git task's file and ownership arguments.
8. Immediately reapply the identical asset-enabled baseline and require the
   aggregate recap to report `changed=0` and `failed=0`.
9. Retain the explicit Seriatim `v1.4.0` forwarding case and its immediate
   clean repeat. Include the same controlled asset inputs so version coverage
   does not switch content sources or fetch mutable asset branches.
10. Ensure best-effort cleanup reapplies the asset-enabled baseline and removes
    only the integration driver's private temporary directory. Do not delete
    the managed campaign or prompt roots, source trees, binaries, runtime
    areas, or any caller content.
11. Update the README validation description only as needed to describe native
    campaign resolution, deployed prompt/schema discovery, controlled local
    asset inputs, clean repetition, and the retained version-forwarding case.
12. Do not add assertions for file ownership, modes, directory existence,
    checkout revisions, remote update mechanics, task order, task results,
    exact current defaults, every campaign, or every prompt. Do not run a real
    backend, WhisperX request, or transcript workflow.

### Validation

From `ansible-role-dnd-pipeline`, run:

```text
bash -n tests/integration
tests/integration
git diff --check
git status --short
```

Expected result: fresh dependency bootstrap, syntax checking, and
production-profile lint pass; controlled local asset revisions deploy through
the real role; Narratio validates a campaign selected from the deployed
registry; Scriptorium loads the repaired prompt and resolves a deployed schema;
the baseline and explicit-version repeats both report `changed=0` and
`failed=0`; cleanup restores the asset-enabled compatible baseline; and only
the intended integration-driver and README changes are present.

### Completion Criteria

- The integration suite uses explicit local Git origins and immutable
  revisions without freezing the role's mutable defaults.
- Narratio proves campaign discovery through its native validator.
- Scriptorium proves prompt discovery through its managed configuration and a
  representative deployed schema resolves under the configured schema root.
- Immediate repeated convergence is clean with the asset layer enabled.
- Existing application-composition and version-forwarding behavior remains
  covered and successful.
- No trusted-module tests, task-trace assertions, broad asset snapshots, or
  unrelated test matrices are added.

### Commit

Commit intent: `Add focused asset integration coverage`

## Final Validation and Handoff

After Stage 4, run:

```text
cd assets/campaigns
git log --oneline -1
git status --short --branch
cd ../narratio-prompts
/usr/local/bin/scriptorium inspect prompt --prompt-dir "$PWD/prompts" --prompt dnd.session_recap --prompt-version 2.0.0 --format json
git log --oneline -1
git status --short --branch
cd ../../ansible-role-dnd-pipeline
tests/integration
git log --oneline -2
git status --short --branch
cd ..
./scripts/status
```

Report the Stage 1 campaign commit, Stage 2 prompt commit, and Stage 3 and
Stage 4 pipeline-role commits; native prompt and campaign validation;
schema-resolution evidence; both clean-repeat results; cleanup outcome; and
every repository's working-tree state. Identify any deviation, unresolved
concern, or complexity beyond the planned three-task production flow and the
acceptance criterion that required it.

Do not edit workspace planning files, mark Feature 0007 `done`, modify
standalone roles, or push during out-of-band implementation. Both asset
commits must eventually be pushed before a default remote deployment can
consume them, but pushing and feature completion remain post-review actions.

## Progress Notes

- [x] Stage 1 complete — `assets/campaigns` commit `d1193e8`; both campaign
  IDs passed native Narratio validation. No deviations.