Writing: read docs/skills/shared/writing-rule.md before you write a durable artifact.

/feature-diagram — Per-feature Mermaid Sub-graph

Reads the Feature Inventory at projects/<name>/feature-inventory.md (produced by /extract-features) and writes a per-feature Mermaid flowchart LR showing the HTTP routes, data models, async jobs, and UI screens that participate in one feature. Output: projects/<name>/features/<slug>.md (one file per feature).

This skill closes the gap left by the rest of the architecture-doc family:

See AgDR-0035-per-feature-diagrams.md for the design rationale.

Path resolution

Read the per-project docs dir via portfolio_projects_dir from .claude/hooks/_lib-portfolio-paths.sh. Source the helper at the top of any bash block that touches that path:

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-read-config.sh"
source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-portfolio-paths.sh"
projects_dir=$(portfolio_projects_dir)

Defaults match today's single-fork layout (./projects). Adopters in split-portfolio mode override the portfolio.{registry, projects_dir} keys in .claude/project-config.json — the helper resolves whichever mode they're in. See docs/multi-project.md.

Write targets (see me2resh/apexyard#373 + #443): paths documented as projects/<name>/X in this skill are canonical adopter-facing forms — implement them in bash as "${projects_dir}/<name>/X". Never construct from "${PWD}/projects/...", "$(git rev-parse --show-toplevel)/projects/...", or a literal ./projects/... — those break in split-portfolio v2 mode where projects_dir resolves to a sibling repo.

REQUIRED per-block preamble (see #443): Claude executes each bash block as a separate shell invocation. The projects_dir assignment from the Path resolution section above does NOT carry into later blocks. Every bash block that writes to a projects/<name>/X path MUST start with this three-line preamble so it's self-contained:

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-read-config.sh"
source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-portfolio-paths.sh"
projects_dir=$(portfolio_projects_dir)
# ... now write to "${projects_dir}/<name>/X"

The Path resolution section's example sources the helper once for documentation purposes; it does not absolve later blocks from sourcing it themselves. Treat each bash fence as a fresh process.

Usage

/feature-diagram create-order                       # current project (cwd inside workspace/<name>/ or single-project fork)
/feature-diagram create-order sample-app            # registered project
/feature-diagram reset-password --force             # overwrite existing diagram

If <project-name> is omitted and cwd is the ops-fork root with multiple registered projects, the skill asks which project's inventory to read.

Output location

projects/<name>/features/<slug>.md                  ← the per-feature diagram (one file per feature)
projects/<name>/feature-inventory.md                ← updated to link the Feature column to the new file

The features/ directory is created if missing.

Activated role

When /feature-diagram runs, activate the Tech Lead role — they own architectural slicing and the "which surfaces participate in this feature" view. See .claude/rules/role-triggers.md for the full activation protocol.

Process

1. Resolve the target

• If <project-name> is given and the registry has it → use that project.
• If no <project-name> and cwd is inside workspace/<name>/ → use that project.
• If no <project-name> and cwd is the ops fork root → ask which registered project (list from apexyard.projects.yaml).
• If exactly one project is registered → use it without asking.

2. Locate the inventory

The inventory file must exist at <projects_dir>/<name>/feature-inventory.md. If missing, stop and tell the operator:

No feature inventory at projects/<name>/feature-inventory.md.
Run /extract-features <project> first — that produces the inventory
this skill reads from.

3. Resolve the feature slug

The slug matches a row in the inventory's consolidated feature matrix. Match in this order:

1. Exact slug match in an existing [link text](features/<slug>.md) cell — re-runs.
2. Feature title match (kebab-case the title and compare) — e.g. operator passes create-order and a row says Create order.
3. Substring match — e.g. password matches Reset password via email. Ambiguous matches (≥ 2 rows) ask the operator to disambiguate.

If no match → exit 2 with a helpful message listing the available slugs:

No feature matches slug 'creat-order' in projects/<name>/feature-inventory.md.

Available slugs:
  create-order
  reset-password-via-email
  bulk-delete-users
  ...

Re-run with one of these, or check the inventory file.

4. Extract the row's surfaces

For the matched row, parse the inventory's per-axis findings tables (HTTP routes, Data models, Async jobs, UI screens) and collect every element whose Notes / File / Source column ties back to this feature.

The inventory's row has a Source column that already names the axes that corroborated the feature (e.g. route + test + UI, model only). Use that as the primary filter:

When the inventory row references specific handlers / models / jobs / components by file path (the per-axis tables in the inventory are keyed by File), follow those references and include each as a node in the matching subgraph.

5. Generate the Mermaid

Build a flowchart LR with four named subgraphs:

flowchart LR
    subgraph Screens["UI Screens"]
        screen_<id>["<Display name><br/>(<file>)"]
        ...
    end
    subgraph Routes["HTTP Routes"]
        route_<id>["<METHOD> <path><br/>(<file>)"]
        ...
    end
    subgraph Models["Data Models"]
        model_<id>["<Name><br/>(<file>)"]
        ...
    end
    subgraph Jobs["Async Jobs"]
        job_<id>["<Name><br/>(<file>)"]
        ...
    end

    %% Edges inferred from the inventory row
    screen_<id> -->|submits| route_<id>
    route_<id> -->|reads/writes| model_<id>
    route_<id> -->|enqueues| job_<id>
    job_<id> -->|reads/writes| model_<id>

Empty subgraphs stay visible — render <Subgraph>_empty["(none)"] as a placeholder so the four-quadrant shape is constant across features:

subgraph Jobs["Async Jobs"]
    Jobs_empty["(none)"]
end

Node IDs are stable: screen_<slug>, route_<slug>, model_<slug>, job_<slug> where <slug> is a kebab-case-ified version of the element's name. Stability matters because operators may hand-add a comment or annotation under a specific node ID; re-runs should keep those grounded.

Edge inference rules — same as documented in AgDR-0035:

Don't invent edges. If the inventory row doesn't corroborate a screen → route mapping (no shared URL prefix, no shared component name), don't draw the edge — emit the nodes only and add a one-line coverage-gap note at the bottom of the file.

6. Write the file

Path: <projects_dir>/<name>/features/<slug>.md. Output template:

# <Feature title>

> Per-feature architectural slice for **<Feature title>**. Generated from the consolidated feature matrix in [`../feature-inventory.md`](../feature-inventory.md).

**Status**: <Active | Deprecated | Untested | Documented but not in code>
**Source axes**: <route + test + UI + job>
**Notes**: <Notes column from the inventory row>

## Diagram

```mermaid
flowchart LR
    ...
```

## Participating elements

### HTTP Routes (<N>)

| Method | Path | Handler | File |
|--------|------|---------|------|
| ... | ... | ... | ... |

### Data Models (<N>)

| Model | Fields | File |
|-------|--------|------|
| ... | ... | ... |

### Async Jobs (<N>)

| Job | Trigger | Handler | File |
|-----|---------|---------|------|
| ... | ... | ... | ... |

### UI Screens (<N>)

| Route | Component | Fields | File |
|-------|-----------|--------|------|
| ... | ... | ... | ... |

## Coverage gaps

- <Any edges the inventory didn't corroborate; any axes that came up empty>

---

_Generated by `/feature-diagram` on YYYY-MM-DD. Re-run when the feature's surfaces change._

Behaviour on existing file:

• File doesn't exist → write directly.
• File exists, no --force → prompt the operator with a default-no offer to overwrite. Print a diff against the proposed content if possible.
• File exists, --force → overwrite without prompting; print a one-line Overwriting <path> to stderr.

7. Update the inventory back-link

In <projects_dir>/<name>/feature-inventory.md's consolidated feature matrix table, replace the matched row's Feature cell with a markdown link:

| 1 | [Create order](features/create-order.md) | API + UI | Active | route + test + UI | POST /api/orders; charges Stripe; sends confirmation email |

Idempotent — if the cell already contains a link to the same target file, leave it. If the cell contains a link to a different target file (rare; manual rename), overwrite.

If the inventory is read-only (no write permission, or the operator passed --no-update-inventory) the back-link step is skipped with a one-line warning.

8. Lint the generated Mermaid

Run lint.sh against the output file:

SKILL_DIR="$(dirname "$(realpath "$0")")"
"$SKILL_DIR/lint.sh" "$out_file" || lint_rc=$?

The lint wraps the shared _lib-mermaid-lint.sh — extracts the ```mermaid flowchart block and validates each via mmdc (mermaid-cli). Graceful-degrades when Node / npx is unavailable (exit 3, advisory only).

Treat exit 1 (parse error) as a hard fail — print the lint output, ask the operator whether to (a) auto-regenerate the offending block, (b) keep the file as-is and fix by hand, or (c) re-run with --skip-lint if mmdc is misbehaving. Exit 3 (Node missing) prints a one-line warning and proceeds.

9. Confirm to the user

✓ <project>: per-feature diagram written

  Diagram: <projects_dir>/<name>/features/<slug>.md
  Inventory back-link: updated row <N> in feature-inventory.md

  Routes: 3   Models: 2   Jobs: 1   Screens: 1
  Mermaid lint: 1 of 1 block parsed cleanly

Preview: open the file on GitHub — Mermaid renders inline.
Re-run /feature-diagram <slug> --force when the feature's surfaces change.

Rules

1. Read-only against the codebase. Never modify the project's source. Only writes to projects/<name>/features/<slug>.md and updates the inventory's back-link.
2. The inventory is a hard dependency. If feature-inventory.md doesn't exist, refuse with a pointer to /extract-features. Don't auto-run it (that's a heavyweight scan; the operator should know they're triggering it).
3. Don't auto-overwrite. Existing diagrams require explicit --force. The diagrams may have been hand-edited.
4. Empty subgraphs stay visible. Four quadrants every time; render (none) placeholders so the reader's eye doesn't have to re-orient between features.
5. Don't invent edges. Every edge must trace to a column in the inventory row. Missing corroboration becomes a coverage-gap note, not a fabricated arrow.
6. Stable node IDs. screen_<slug> / route_<slug> / model_<slug> / job_<slug> — slug is kebab-case-ified element name. Reproducibility matters; operators may annotate by ID.
7. Footer signature is mandatory. Every generated file ends with the Generated by /feature-diagram on YYYY-MM-DD line so future readers know it's regenerable.
8. Refuse if the slug doesn't match a row. Exit 2 with the list of available slugs — no silent best-guess fallback.

When to use this

Out of scope (v1)

• Cross-feature diagrams (showing how features depend on each other) — listed as out-of-scope in #288.
• Per-feature sequence diagrams — use templates/architecture/sequence.md manually if needed.
• Trust-boundary overlays — that's /dfd's slice. If the operator wants security-scoped feature analysis, run /dfd and /feature-diagram side by side.
• Auto-running /extract-features — the inventory is a hard dependency, not an optional auto-trigger.
• Cross-project per-feature diagrams — single project per invocation. Multi-project per-feature is a separate concern.

Anti-patterns

• Don't substitute /feature-diagram for /c4. The per-feature slice shows ONE feature's surfaces. System-wide topology lives in /c4.
• Don't run /feature-diagram for every row on every PR. It's refresh-on-arch-change, not per-PR. The whole inventory only changes when features ship; the per-feature diagrams only change when the feature's routes / models / jobs / screens change.
• Don't hand-edit a generated diagram and lose the footer. Re-runs detect the footer; without it the skill can't tell if a file is regenerable. If you must hand-edit, keep the footer line in place.
• Don't treat coverage gaps as failures. The inventory has gaps by design (the scanner can't see business rules, permission matrices, or implicit features). Per-feature diagrams inherit those gaps — that's a feature, not a bug; the gaps are surfaced explicitly so a human can fill them in.

Part of ApexYard — multi-project SDLC framework for Claude Code · MIT.