runtime-attempt-report

Purpose

Record a chronological, secret-free runtime attempt log—including version stamps, downgrade-only truth labels, and routed next steps—for Path 1, Path 2, or CLI appendix runs. A report produced by this skill can serve as an in-repo provenance record under docs/evals/, without certifying artifact quality or diagnosing runtime defects.

Quick start

• gather timestamps, commands/config surfaces touched (no secrets)
• choose blocks from references/truth-label-blocks.md
• fill references/attempt-record-template.md
• route improvements via references/triage-next-steps.md

When to use

• after any Blender connect attempt (success or failure)
• preparing notes for docs/evals/ style provenance
• teaching non-technical stakeholders what actually occurred

When not to use

• Classifying rendered or exported visual artifacts (render-export-evidence) → use ../render-export-evidence/SKILL.md (render-export-evidence)
• Diagnosing systematic runtime or scene defects (blender-troubleshooting) → use ../blender-troubleshooting/SKILL.md (blender-troubleshooting)
• Running pre-attempt environment checks (runtime-readiness) → use ../runtime-readiness/SKILL.md (runtime-readiness)
• Selecting which runtime path to use before an attempt (runtime-path-picker) → use ../runtime-path-picker/SKILL.md (runtime-path-picker)
• Performing final release or delivery gating (pre-handoff-verification) → use ../pre-handoff-verification/SKILL.md (pre-handoff-verification)

Trigger phrases

• "write up what we tried"
• "runtime attempt log"
• "postmortem blender connect"
• "evidence report template"

Prerequisites / readiness

• Pre-attempt path selection completed via runtime-path-picker and readiness checked via runtime-readiness
• Chronological list of actions attempted (even if partial or aborted)
• Exact version stamps for Blender, MCP host/client, and any bridge/add-on used

Input schema

Required inputs

Optional inputs

Output schema

Primary output

• Structured Markdown attempt report (sections A–F) containing chronological actions, redacted config surfaces, version stamps, truth labels, and routed next steps (suitable for saving under docs/evals/).

Evidence / caveat output

• Runtime truth state (exact allowed values): Not Run | Attempted | Produced | Verified | Failed
• Artifact truth state (default Not Produced; exact allowed values): Not Run | Not Produced | Produced | Verified | Failed
• Discipline: Downgrade-only, never upgrade; defer all artifact verification and visual quality ledgering to render-export-evidence.
• Embedded template includes explicit Runtime + Artifact states.

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

Documentation skill only—never executes Blender or generates Python. Reports must strictly preserve upstream runtime distinctions:

• Path 1: Official Blender Lab MCP in Blender 5.1+, hosted from (a) Anthropic Connector or (b) manual MCP client. The Connector is NOT standalone.
• Path 2: Community mcp-for-blender (third-party, Blender 3.0+); document only with explicit caveats and never treat as release evidence.
• CLI fallback: Appendix only (blender --background --python); remains upstream-documented with no in-repo evidence file unless a report written by this skill becomes one under docs/evals/.

Operating procedure

1. Stamp the attempted path, host mode, and exact versions (Blender, host/client, add-on/bridge).
2. Record operator actions chronologically in plain language with all secrets, tokens, and private paths redacted.
3. Assign Runtime (Not Run | Attempted | Produced | Verified | Failed) and Artifact (default Not Produced; allowed Not Run | Not Produced | Produced | Verified | Failed) truth states using strict downgrade-only discipline.
4. Route next steps: send generated files to render-export-evidence, defect symptoms to blender-troubleshooting or bridge-conflicts, and delivery readiness to pre-handoff-verification.

Decision tree

Did a real runtime connect or CLI execution attempt actually start?
  No (environment blocked before start, or dry-run walkthrough) →
    Set Runtime status: Not Run and Artifact status: Not Produced; record pre-attempt blocker and route to runtime-readiness
  Yes → continue
Which runtime path was attempted (`references/attempt-record-template.md`)?
  Path 1 (Official Blender Lab MCP in Blender 5.1+, Host (a) Connector or Host (b) manual MCP client) →
    Stamp Blender 5.1+, Lab add-on, and host versions; note Connector is not standalone
  Path 2 (Community `mcp-for-blender`, Blender 3.0+) →
    Stamp versions, attach mandatory third-party / non-release-evidence caveat (`docs/unofficial-runtime-bridges.md`)
  CLI appendix (`blender --background --python`) →
    Stamp Blender version, redacted command flags, and stdout/stderr pointers
How did the attempt conclude (`references/truth-label-blocks.md`)?
  Aborted mid-flight or timed out without conclusive outcome → Runtime status: Attempted
  Terminated with explicit error / crash → Runtime status: Failed; route to blender-troubleshooting or bridge-conflicts
  Completed with observable runtime signal → Runtime status: Produced (or Verified only if smoke checks are recorded)
Were any render/export files written to disk?
  No → Artifact status: Not Produced
  Yes → cap Artifact status at Produced (never Verified here) and route files to render-export-evidence

Playbooks

Playbook A: Path 1 read-only or scene-mutation attempt log

• Populate Sections A–F of references/attempt-record-template.md with exact timestamps, Blender 5.1+ version, Lab MCP add-on version, and Host (a) (Anthropic Connector) or Host (b) (manual MCP client) version stamps.
• Redact local user paths, tokens, and environment secrets before logging chronological actions.
• Assign downgrade-only truth states from references/truth-label-blocks.md and route any produced files to ../render-export-evidence/SKILL.md.

Playbook B: Failed or timed-out MCP connection postmortem

• Record the exact chronological step where the connection dropped or port 9876 refused attachment (references/triage-next-steps.md).
• Set Runtime status: Failed (or Attempted if inconclusive) and Artifact status: Not Produced — never invent a Blocked truth state.
• Route port/bridge collisions to ../bridge-conflicts/SKILL.md, runtime errors to ../blender-troubleshooting/SKILL.md, and handoff gating to ../pre-handoff-verification/SKILL.md.

Playbook C: Path 2 community bridge or CLI appendix provenance record

• For Path 2 (mcp-for-blender), stamp versions and include the explicit disclaimer that third-party bridge runs never count as official release evidence.
• For CLI appendix (blender --background --python), log the phased command invocation and stdout/stderr tail so the report can serve as an in-repo provenance candidate under docs/evals/.

Mode handling

Text-only mode

• Record hypothetical or dry-run walkthroughs with Runtime set to Not Run and Artifacts set to Not Produced. Never upgrade states without a real attempt.

Runtime-ready mode

• Log chronological steps and partial transcripts; cap Runtime at Attempted or Produced unless independently verified, and keep Artifacts at Not Produced or Produced pending render-export-evidence.

Blocked runtime mode

• Document the exact blocker chronologically, set Runtime to Not Run or Failed (downgrade-only), keep Artifacts at Not Produced, and route to runtime-readiness or pre-handoff-verification.

Validation checklist

• Runtime state uses only Not Run | Attempted | Produced | Verified | Failed and Artifact state defaults to Not Produced (Not Run | Not Produced | Produced | Verified | Failed)
• Truth labels follow downgrade-only discipline (no state upgrades, no Path 2 release evidence claims)
• Blender, host/client, and bridge/add-on version stamps are recorded
• Chronological actions contain no secrets, credentials, or upstream vendor blame language
• Operator sandbox hygiene recorded (trusted environment, no credentials in scenes/logs, disposable ports/paths)
• Artifact inspection is delegated to render-export-evidence and handoff gating is routed to pre-handoff-verification

Pass / Warn / Fail rubric

Failure handling

• If an attempt aborts or crashes mid-run, record the chronological steps up to the failure point and set Runtime state to Failed and Artifact state to Not Produced (or Failed if corrupted).
• If version stamps or chronological steps cannot be confirmed from operator notes, downgrade the Runtime state to Attempted (or Not Run) and flag the gap explicitly.
• If a report is generated for a Path 2 (mcp-for-blender) run, attach the mandatory third-party caveat and block it from serving as release evidence.
• Route any failed or incomplete attempt report directly to pre-handoff-verification to block downstream delivery until resolved.

Troubleshooting

• Missing version metadata: Do not guess versions; mark the version field as unrecorded, downgrade truth claims accordingly, and route to runtime-readiness to capture environment facts.
• Ambiguous connection drops: Log the exact timestamp and symptom chronologically without speculating on root causes, then hand off diagnosis to blender-troubleshooting or bridge-conflicts.
• Unverified output files present on disk: Keep Artifact state at Produced (never Verified) and route the files to render-export-evidence for inspection.
• Sensitive tokens in raw logs: Strip all credentials, API keys, and personal file paths before writing the chronological record.

Best practices

• Stamp exact versions for Blender, the MCP host/client, and any bridge add-on at the top of every report.
• Write chronological entries in plain, factual language suitable for non-Blender stakeholders and docs/evals/ provenance records.
• Default Artifact truth state to Not Produced and apply downgrade-only logic whenever logs or outcomes are partial.
• Clearly distinguish Path 1, Path 2, and CLI appendix runs in every report.

Good examples

• Path 1 manual MCP client attempt: Logs Blender 5.1+ and client version stamps, chronological connection steps, Runtime status: Produced, Artifact status: Not Produced, and routes next steps to render-export-evidence.
• Blocked Path 1 Anthropic Connector attempt: Notes that the Connector was invoked without a running Blender 5.1+ Lab add-on host (Connector is not standalone), stamps Runtime status: Failed and Artifact status: Not Produced, and routes to runtime-readiness and pre-handoff-verification.
• Path 2 community exploration log: Records a Blender 3.0+ mcp-for-blender run with explicit third-party caveats, caps truth labels against release use, and routes handoff gating to pre-handoff-verification.
• CLI appendix record for docs/evals/: Documents a blender --background --python run chronologically with redacted paths and version stamps to establish an in-repo evidence file.

Bad examples

• Claiming Runtime status: Verified and Artifact status: Verified from partial console logs or operator optimism without running render-export-evidence.
• Pasting raw MCP configuration files containing API keys, bearer tokens, or unredacted personal filesystem paths into Section C or Section E.
• Presenting a third-party Path 2 (mcp-for-blender) session transcript as official Draft v0 release certification.

User-facing response template

## A. Summary
Path: <Path 1 Official Blender Lab MCP (5.1+) | Path 2 community mcp-for-blender (3.0+, non-release) | CLI fallback appendix>
Host: <Host (a) Anthropic Connector | Host (b) Manual MCP client | Headless shell>
Date/time (UTC): <timestamp>

## B. Versions
Blender: <exact version or unrecorded>
Add-on / bridge package: <Lab MCP add-on version | mcp-for-blender version | N/A>
MCP host / client: <client name + version>

## C. Actions chronology (secrets & private paths redacted)
1. <timestamp — action taken>
2. <timestamp — action taken>

## D. Signals observed
- Expected: <observed expected signals>
- Unexpected: <timeouts, port 9876 conflicts, or errors>

## E. Truth labels (downgrade-only)
Runtime status: <Not Run | Attempted | Produced | Verified | Failed>
Artifact status: <Not Produced | Produced | Verified | Failed>
Evidence pointers: <redacted log/file paths or none>

## F. Limitations, redactions & next routing
Secrets removed: Yes (trusted sandbox hygiene verified)
Limitations: <missing version stamps, third-party Path 2 caveat, or unverified outputs>
Next skill: <render-export-evidence | blender-troubleshooting | bridge-conflicts | pre-handoff-verification>

Anti-patterns

• Upgrading Runtime or Artifact states to Verified based on partial console logs or operator optimism.
• Writing Artifact status: Not Run instead of Artifact status: Not Produced (or inventing a Blocked truth state).
• Treating a Path 2 (mcp-for-blender) attempt report as release certification or official runtime evidence.
• Blurring roles by inspecting render quality inside the attempt report instead of routing to render-export-evidence, or diagnosing defects instead of routing to blender-troubleshooting.
• Pasting raw config files containing API keys, tokens, or local user credentials into the report.

Cross-skill handoff

• [upstream_reference] Pre-attempt readiness check → ../runtime-readiness/SKILL.md
• [upstream_reference] Runtime path selection context → ../runtime-path-picker/SKILL.md
• [fallback] Bridge / port / host conflicts → ../bridge-conflicts/SKILL.md
• [fallback] Runtime or scene defect diagnosis → ../blender-troubleshooting/SKILL.md
• [progression] Artifact evidence classification → ../render-export-evidence/SKILL.md
• [fallback] Final handoff gate & failure exit → ../pre-handoff-verification/SKILL.md

Non-goals

• Executing Blender, running MCP commands, or writing Python scripts.
• Certifying visual artifact quality or export integrity (owned by render-export-evidence).
• Diagnosing runtime crashes, bridge conflicts, or scene errors (owned by blender-troubleshooting and bridge-conflicts).
• Approving final delivery packages without passing through pre-handoff-verification.

References

• references/truth-label-blocks.md
• references/attempt-record-template.md
• references/triage-next-steps.md
• ../../laws/evidence-before-done.md