--- name: cli-appendix-job-outline description: Plan headless blender --background export or render jobs when MCP chat control is a poor fit; cite upstream Blender CLI docs only. --- # cli-appendix-job-outline ## Purpose Structure **CLI fallback appendix** jobs (batch render, batch export, scripted scene setup) using Blender's documented command-line interface — without presenting arbitrary Python as the user-facing product surface. ## Quick start - classify job type via `references/job-type-matrix.md` - capture ordering cautions `references/flag-order-reminders.md` - state limitations `references/stateless-limitations.md` - point user to upstream CLI manual for exact flags ## When to use - CI / render farm / repeatable exports - MCP unavailable or disallowed by policy - deterministic batch operations vs exploratory modeling ## When not to use - interactive viewport iteration primary goal - user expects Connector chat loop ## Trigger phrases - "headless blender" - "batch render CLI" - "blender --background" - "render farm outline" ## Prerequisites / readiness - user accepts stateless job model - Blender executable accessible to shell running job ## Input schema ### Required inputs | Input | Why it is required | |---|---| | Job intent (batch render vs batch export vs scripted job) | Chooses matrix row (routeMCP vs CLI selection to `runtime-path-picker` first) | | Output artifact expectations | Drives validation hooks (route format selection to `output-format-decision`) | ### Optional inputs | Input | Use | |---|---| | GPU availability | Device flags category note | | Mixture of Windows/Linux paths | Quote escaping caution | ## Output schema ### Primary output - Phased job outline: startup, load, process, save - Flag categories to look up in the upstream Blender manual (plain language only; no copy-paste Python dumps, no execution) - Validation and logging hooks (stdout tail, expected output paths) ### Evidence / caveat output ```txt Runtime status: Not Run | Attempted | Produced | Verified | Failed Artifact status: Not Produced | Produced | Verified | Failed (default: Not Produced) Rule: Downgrade-only, never upgrade. Outlining a job never executes Blender (Runtime remains Not Run; Artifact remains Not Produced). Blockers: Evidence used: Limitations: Stateless execution, no interactive viewport ``` ## Required laws - `../../laws/evidence-before-done.md` - `../../laws/non-blender-user-language.md` - `../../laws/no-arbitrary-python-interface.md` - `../../laws/official-runtime-only.md` ## Official runtime boundary Planning outline only — BlendOps never executes Blender and never presents arbitrary Python as the product surface. This CLI workflow is an **appendix only**, never a peer path without an eval record, and BlendOps has **no in-repo CLI evidence file yet**. Point users to the upstream Blender manual for exact CLI flags. ## Operating procedure 1. Verify `runtime-path-picker` confirmed MCP is ill-fit; if interactive MCP is needed, hand off to `path-one-host-outline`. 2. Verify target formats are settled via `output-format-decision`. 3. Select the job-type matrix row and outline phases (`blender --background` startup, load, process, save), flag categories, and validation/logging hooks in plain language without Python dumps. 4. Direct the user to verify exact flags against the upstream Blender manual. 5. Keep truth states at `Not Run` (Runtime) and `Not Produced` (Artifact); hand off actual CLI execution logs and outputs to `render-export-evidence`. ## Decision tree ```txt Does the user need interactive viewport look-dev or live chat steering? Yes → do NOT use CLI appendix; route to runtime-path-picker or path-one-host-outline No → continue Has MCP vs CLI path selection already been decided? No → route to runtime-path-picker first to confirm CLI appendix fits constraints Yes → continue Are target output formats and resolutions settled? No → route to output-format-decision / resolution-aspect-decision before outlining batch stages Yes → continue Which batch job pattern applies (`references/job-type-matrix.md`)? Batch still render → outline startup, .blend load, engine/camera/output flag categories, and render log hooks Batch GLB export → outline startup, .blend load, collection visibility/export check categories, and GLB size/validator hooks Scripted scene hygiene / prep → outline factory-startup caution, `--python` ordering reminder, and file diff/hash hooks Did the operator already run the CLI job and produce logs or files? Yes → route logs to runtime-attempt-report-writer and output files to render-export-evidence No → emit planning outline with Runtime status: Not Run and Artifact status: Not Produced ``` ## Playbooks ### Playbook A: CI or render-farm headless still render outline - Select the **Batch still render** row from `references/job-type-matrix.md`. - Structure the 4 phases (background startup, scene file load, render execution flags category, output verification hooks) and attach flag-ordering cautions from `references/flag-order-reminders.md`. - Direct the operator to the upstream Blender CLI manual for exact version-matched flags and keep `Runtime status: Not Run` and `Artifact status: Not Produced`. ### Playbook B: Nightly headless GLB batch export outline - Select the **Batch GLB export** row from `references/job-type-matrix.md` after confirming format rules via `output-format-decision`. - Highlight stateless limitations (`references/stateless-limitations.md`), extension/add-on enablement differences under factory startup profiles, and argument-separator (`--`) cautions. - Specify downstream evidence hooks for `render-export-evidence` and `glb-web-handoff`. ### Playbook C: Restricted IT environment where MCP ports are blocked - Acknowledge the policy blocker on a separate `Blockers:` line when MCP bridges (port `9876`) are disallowed. - Frame the `blender --background` workflow strictly as the **CLI fallback appendix** (documented upstream as a first-class Blender CLI surface, with no in-repo BlendOps eval record yet). - Provide the phased job outline and logging hooks without pasting executable Python scripts. ## Mode handling ### Text-only mode Provide the phased CLI job outline, flag categories, and stateless limitations. Runtime is `Not Run`; Artifact is `Not Produced`. ### Runtime-ready mode Specify validation and logging hooks (where logs and output files should land), then hand off post-run verification to `render-export-evidence` and `runtime-attempt-report-writer`. Outlining never upgrades truth states. ### Blocked runtime mode Outline the intended phased job structure and record blockers on a separate line with Runtime `Not Run` and Artifact `Not Produced`. ## Validation checklist * [ ] Framed strictly as an appendix, never a peer path without an eval record * [ ] Stateless, no-viewport limitations acknowledged * [ ] Upstream Blender CLI manual cited for exact flags * [ ] No copy-paste Python dumps and no implication that BlendOps executes Blender * [ ] Truth states respect downgrade-only rule (default `Not Produced` for artifacts; outlining never runs anything) * [ ] Role boundaries preserved with `runtime-path-picker`, `output-format-decision`, and `render-export-evidence` ## Pass / Warn / Fail rubric | Verdict | Criteria | |---|---| | Pass | Coherent phased outline + flag categories + validation/logging hooks + accurate downgrade-only truth states | | Warn | GPU or OS path environment uncertain — dual-branch caution noted | | Fail | Includes copy-paste Python dumps, treats CLI as a peer path, blurs roles with path/format/evidence skills, or upgrades truth states without verified evidence | ## Failure handling - If the user asks for a copy-paste Python script or raw `bpy` batch code, refuse the code dump per `laws/no-arbitrary-python-interface.md` and provide a plain-language phased job outline plus upstream manual pointers instead. - If the user wants live interactive viewport adjustments through the CLI appendix, stop and route back to `runtime-path-picker` or `path-one-host-outline`. - If a prior CLI run failed silently (for example, flags placed after the scene filename were ignored before load), keep `Runtime status: Failed` (or `Attempted`), keep `Artifact status: Not Produced`, and route the log capture to `runtime-attempt-report-writer`. - Never promote the CLI fallback appendix to a peer MCP runtime path when no in-repo CLI evaluation record exists under `docs/evals/`. ## Troubleshooting | Symptom | Cause | Safe resolution | |---|---|---| | CLI job renders default cube or wrong settings | Flag order error in `blender --background` (settings placed after render trigger or scene file) | Review `references/flag-order-reminders.md` and verify exact flag ordering in the upstream Blender command-line manual for the installed Blender version. | | Custom script arguments disappear in headless run | Missing `--` argument separator after `--python` script path | Note the argument-separator check prompt from `references/flag-order-reminders.md` and consult upstream CLI docs. | | Export add-on or GPU device missing on CI runner | Factory startup or headless runner profile differs from artist workstation | Flag the environment gap under `Limitations` (`references/stateless-limitations.md`) and require stdout/stderr device capture in `runtime-attempt-report-writer`. | ## Best practices - Always label `blender --background` workflows as the **CLI fallback appendix** — documented upstream as a first-class Blender CLI surface, while noting BlendOps has **no in-repo CLI evidence file yet**. - Express job steps as phased operational checkpoints (`Startup → Load → Process → Save & Verify`) rather than raw shell one-liners or Python code blocks. - Remind operators that argument order in Blender's CLI is positional and must be verified against the exact Blender manual version running on the host. - Pair every CLI outline with explicit stdout/stderr log paths and output file checks consumed by `render-export-evidence`. ## Good examples - "CLI appendix job outline (Batch GLB export): Phase 1 Background startup (headless mode, verify extension profile) → Phase 2 Load `.blend` source → Phase 3 Trigger GLB export step with `--` argument boundary checked against upstream Blender CLI manual → Phase 4 Capture exit code, stdout/stderr log path, and exported `.glb` byte size for `render-export-evidence`. Runtime status: Not Run. Artifact status: Not Produced." - "Because interactive viewport look-dev is required, `cli-appendix-job-outline` is not the right surface (`references/stateless-limitations.md`); hand off to `runtime-path-picker` to evaluate Path 1 (Blender 5.1+ Lab MCP) or Path 2 (`mcp-for-blender`)." ## Bad examples - "Run this 120-line Python script with `blender -b scene.blend -P export_all.py` and your GLBs are Verified." — violates `laws/no-arbitrary-python-interface.md` and `laws/evidence-before-done.md`. - "Path 3 (Blender CLI) is a verified peer runtime in BlendOps." — misrepresents the CLI appendix as a peer path and fabricates in-repo eval evidence. - " Blender 5.1+ is required to run `blender --background`." — falsely applies Path 1's Lab MCP add-on version floor to the CLI appendix. ## User-facing response template ```txt Job classification: (CLI fallback appendix — not a peer MCP path; no in-repo BlendOps CLI evidence file yet) Upstream reference: Verify exact positional flags in the official Blender Command Line Manual for your Blender version. Phased job outline: 1. Startup: 2. Load: 3. Process: 4. Save & logging hooks: Stateless limitations: Blockers: Runtime status: Not Run Artifact status: Not Produced Evidence used: Next skill: ``` ## Anti-patterns - Dumping copy-paste Python (`bpy`) scripts or claiming BlendOps ships a CLI wrapper binary. - Presenting the CLI fallback appendix as a peer path to Path 1 / Path 2 without an in-repo evaluation record under `docs/evals/`. - Applying the Path 1 **Blender 5.1+** floor (`blender_version_min = "5.1.0"`) to `blender --background` CLI jobs. - Writing `Artifact status: Not Run` or upgrading `Runtime status` / `Artifact status` merely from outlining a batch job. - Skipping `output-format-decision` or `render-export-evidence` when planning deliverable verification. ## Cross-skill handoff * Pick MCP vs CLI → `../runtime-path-picker/SKILL.md` * Interactive MCP host outline → `../path-one-host-outline/SKILL.md` * Export format choices → `../output-format-decision/SKILL.md` * Actual CLI output evidence recording → `../render-export-evidence/SKILL.md` * Runtime attempt reporting → `../runtime-attempt-report-writer/SKILL.md` ## Non-goals - Generating or debugging user-facing Python (`bpy`) scripts. - Executing `blender --background` jobs or inspecting output files directly. - Selecting between MCP Path 1, Path 2, and CLI (owned by `runtime-path-picker`). - Certifying rendered frames or exported GLBs (owned by `render-export-evidence`). ## References - `references/job-type-matrix.md` - `references/flag-order-reminders.md` - `references/stateless-limitations.md` - `../../docs/runtime-stack-strategy.md`