Update to LinkedIn Animated Infographics

{
  "description": "Run LinkedIn infographic work in autopilot mode by negotiating available host capabilities, using real side jobs and sandbox artifacts when available, and falling back honestly when they are not. Use for end-to-end creation, redesign, animation, or rigorous production workflows in ChatGPT or Codex.",
  "included_files": [
    {
      "relative_path": "references/artifact-workspace.md",
      "size_in_bytes": 2288
    },
    {
      "relative_path": "references/asset-source-policy.md",
      "size_in_bytes": 4023
    },
    {
      "relative_path": "references/autopilot-failure-policy.md",
      "size_in_bytes": 1955
    },
    {
      "relative_path": "references/capability-negotiation.md",
      "size_in_bytes": 2281
    },
    {
      "relative_path": "references/execution-paths.md",
      "size_in_bytes": 3964
    },
    {
      "relative_path": "references/narrative-taste.md",
      "size_in_bytes": 5481
    },
    {
      "relative_path": "references/side-jobs.md",
      "size_in_bytes": 3637
    },
    {
      "relative_path": "references/tool-usage-policy.md",
      "size_in_bytes": 2090
    },
    {
      "relative_path": "references/visual-intelligence-capsule.json",
      "size_in_bytes": 44528
    },
    {
      "relative_path": "references/visual-quality-contract.md",
      "size_in_bytes": 6468
    },
    {
      "relative_path": "references/workspace-agents-bridge.md",
      "size_in_bytes": 1568
    },
    {
      "relative_path": "scripts/autopilot_runtime.py",
      "size_in_bytes": 5165
    }
  ],
  "name": "linkedin-infographic-autopilot",
  "skill_md_contents": "---\nname: linkedin-infographic-autopilot\ndescription: Run LinkedIn infographic work in autopilot mode by negotiating available host capabilities, using real side jobs and sandbox artifacts when available, and falling back honestly when they are not. Use for end-to-end creation, redesign, animation, or rigorous production workflows in ChatGPT or Codex.\n---\n\n# LinkedIn Infographic Autopilot\n\n## Purpose\n\nOperate as the parent production workflow for LinkedIn infographic work. Use the strongest capabilities actually exposed by the current host without inventing agents, tools, renders, background work, publishing actions, asset downloads, or reference inspections.\n\nQuality parity means the same discipline as a strong multi-role creative workflow while allowing a different visual direction for every host and every brief.\n\nRead these local references before execution:\n\n- `references/capability-negotiation.md`\n- `references/execution-paths.md`\n- `references/side-jobs.md`\n- `references/artifact-workspace.md`\n- `references/tool-usage-policy.md`\n- `references/autopilot-failure-policy.md`\n- `references/workspace-agents-bridge.md`\n- `references/asset-source-policy.md`\n- `references/narrative-taste.md`\n- `references/visual-quality-contract.md`\n- `references/visual-intelligence-capsule.json`\n\nRead the generated visual-intelligence capsule before creative selection. Apply its hard filters, fixed weights, and slug tie-break, then pass only selected mechanism guidance into production. With no visual reference, reference diagnosis is `SKIP`. Explicit reference intent with evidence that cannot be inspected is `HOLD: reference evidence unavailable` before evidence research. Persistent reference ingestion and source reference media are unavailable in the installed distribution; never claim they were loaded. The capsule is abstract local guidance only.\n\nWhen repository access actually exists, `demos/` may be inspected as a bounded taste corpus through the contract in `references/narrative-taste.md`. Installed skills do not pretend those demo media are present locally.\n\nThe complete blocking visual gate in `references/visual-quality-contract.md` is authoritative for Autopilot. The summary below never replaces or weakens it.\n\n## Autopilot rule\n\nStart every substantial task by observing the capabilities the host actually exposes. Unknown capabilities are unavailable until observed.\n\nSelect exactly one runtime path:\n\n- `full-autopilot`\n- `tool-rich-sequential`\n- `safe-skill-only`\n\nPrefer the strongest safe path. Never pretend that a stronger path ran.\n\n## Executable runtime helper\n\nWhen `shell_or_code_execution` is observed, use `scripts/autopilot_runtime.py` to make orchestration decisions deterministic instead of relying only on free-form model judgment.\n\nPass only capabilities actually observed in the current host. Example shape:\n\n```json\n{\n  \"subagents\": true,\n  \"sandbox_write\": true,\n  \"shell_or_code_execution\": true,\n  \"image_inspection\": true,\n  \"web_research\": true\n}\n```\n\nThe helper can normalize unknown values to unavailable, select the execution path, generate a bounded evidence-first side-job dispatch plan, and initialize the logical artifact workspace without fabricating artifact files.\n\nTypical commands when execution is available:\n\n```text\npython scripts/autopilot_runtime.py select-path --capabilities-json '<observed-json>'\npython scripts/autopilot_runtime.py dispatch-plan --capabilities-json '<observed-json>'\npython scripts/autopilot_runtime.py init-workspace <task-workspace>/work\n```\n\nIf code execution is not observed, apply the same contracts directly from the references. Do not claim the helper ran.\n\n## Identity behavior\n\nFor every named logo or mascot, apply `references/asset-source-policy.md` before creative direction.\n\nIdentity precedence is exact user-supplied official asset, inspectable original-owner source, pinned Vibe SVGs logo mirror for platform/tool logos, verified Lobe fallback when appropriate, then HOLD.\n\n`https://github.com/imMamdouhaboammar/vibe-svgs` is a useful curated source/discovery surface, but its `communityArtwork: true` mascot/scene assets are fan-made/community artwork and must never be promoted to official identity. Official/original mascot requests require exact user/original-owner provenance or HOLD.\n\nEvery approved identity becomes identity-locked. Do not redraw, trace, reconstruct, recolor without an approved source variant, distort aspect ratio, or alter wordmark/character-defining geometry.\n\n## Reference and narrative behavior\n\nWhen inspectable visual references are supplied, derive reusable guidance rather than copying the reference. For every selected mechanism use:\n\n`Evidence -> Observation -> Transferable Rule -> Anti-Rule`\n\nWhen multiple references exist, assign explicit non-overlapping jobs such as hook, progression, composition, type hierarchy, color, texture, proof, pacing, motion, or payoff. Do not blend them into one vague style and do not copy distinctive subject matter, identity, proprietary artwork, or unique composition signatures.\n\nBefore macro layout, apply `references/narrative-taste.md`. Build a real visual progression, commonly `Hook -> Tension -> Evidence -> Turn -> Payoff` when it fits the evidence. Every retained beat records a Reader question and Beat-to-visual mapping. A list converted into cards is not a sufficient narrative.\n\n## Required production sequence\n\n1. Observe runtime capabilities and select the execution path.\n2. Run evidence research/inventory first and finalize `evidence/evidence.json` or its non-materialized equivalent.\n3. Resolve every required named logo/mascot through `references/asset-source-policy.md`; localize/record provenance when execution allows it. HOLD on unresolved official identity.\n4. Only after evidence and required identity boundaries are finalized, fan out independent creative discovery side jobs when real delegation is observed.\n5. Generate at least three structurally different creative directions from the finalized evidence boundary, applying reference-transfer rules when references exist.\n6. Select the direction and define the primary takeaway.\n7. Build the narrative contract from `references/narrative-taste.md`: story shape, hook, ordered beats, Reader question, evidence dependency, Beat-to-visual mapping, turn/payoff, persistent context, reference/demo jobs, anti-rules, originality changes, and necessary motion jobs.\n8. Compress copy into visual slots using only approved evidence and narrative beats.\n9. Define macro layout before component styling, preserving approved identity assets and persistent narrative context.\n10. Run the complete blocking perception preflight from `references/visual-quality-contract.md` and repair only the smallest responsible dimension when a check fails.\n11. Build the still when sandbox writes are observed.\n12. Run the complete still critique taxonomy, identity-integrity checks, perception tests, severity pressure, and targeted revision routing.\n13. Repair blocking still defects, maximum two targeted repair attempts.\n14. Proceed to motion only after the complete still gate passes.\n15. Implement motion only when the requested output and host capabilities support it; each motion advances reading order, state change, travel, or the narrative turn/payoff.\n16. Run render QA using actual rendered evidence when image inspection is observed, including identity integrity and narrative readability across sampled states.\n17. Run an independent final verification pass.\n18. Deliver artifacts plus a truthful capability/execution summary.\n\n## Side jobs\n\nWhen real child-agent or equivalent side-job execution is observed, use dependency-aware delegation rather than making the parent do everything serially.\n\nEvidence and required identity resolution are not part of the first creative fan-out. They must complete first because every downstream creative job consumes their finalized boundaries.\n\nAfter those boundaries are finalized, the creative fan-out may run independent jobs in parallel:\n\n- creative direction exploration\n- visual archetype/narrative-shape exploration\n- copy compression critique\n\nThe parent waits for required side jobs, validates bounded artifacts, selects what survives, and owns the final decision.\n\nWhen real delegation is not observed, run the same contracts sequentially in the same dependency order. Do not describe sequential role passes as agents.\n\n## Sandbox and artifact behavior\n\nWhen sandbox or workspace writes are observed, persist intermediate work using the logical artifact contract in `references/artifact-workspace.md`.\n\nUse artifacts to reduce transient-context loss and to make QA inspect the same inputs that the build used.\n\nWhen write access is unavailable, do not claim HTML, images, GIFs, screenshots, downloaded assets, or files were created. Produce the strongest truthful planning/critique output and return `HOLD` when the requested final artifact cannot be executed.\n\n## Tool behavior\n\nUse tools, apps, search, code execution, image inspection, and publishing connectors only when they materially improve a named job.\n\nRecord important tool results in the artifact or verification stage that consumed them.\n\nA tool being described in documentation is not proof that it exists in the current session. Capability must be observed.\n\n## Visual gates\n\nApply every rule and every PASS/FAIL taxonomy item in `references/visual-quality-contract.md`.\n\nFor the default LinkedIn format the gate includes:\n\n- 1080x1350 canvas unless the user requests another format\n- roughly 82-92% usable vertical occupancy unless intentional negative space has a clear compositional job\n- rejection of unexplained footer/final-zone gaps greater than 120px\n- maximum bordered containment depth of two levels\n- one dominant visual anchor at feed scale\n- explicit macro rhythm and footer reservation\n- blocking perception preflight including one-second hierarchy, 100x100 thumbnail, squint/blur, grayscale, negative-space, tangency, brand-off specificity, and effect-subtraction checks\n- severity-aware visual-slop pressure that never cancels a hard-gate failure\n- rejection of generic UI grammar, nested-card density, weak macro rhythm, weak visual anchor, top-heavy composition, and feed-scale legibility failure\n- blocking identity drift for named logos/mascots\n- blocking narrative failure when beats do not change understanding or motion merely decorates a weak still\n\nThe critic must explicitly return PASS or FAIL for the full taxonomy, including `top-heavy-composition`, `bottom-dead-zone`, `nested-card-density`, `generic-ui-grammar`, `weak-macro-rhythm`, `weak-visual-anchor`, `footer-detachment`, `motion-on-weak-still`, `decorative-motion`, and `feed-scale-legibility`.\n\nMotion cannot begin while any blocking still item or perception-preflight item fails.\n\nIf a blocking still defect remains after two targeted repairs, return `FAIL:fixable` or `HOLD` instead of shipping weak output.\n\n## Focused capability routing\n\nWhen the host exposes the related public skill and the request is focused, prefer the narrower contract:\n\n- `linkedin-infographic-studio` for full production without explicit autopilot orchestration\n- `linkedin-infographic-review` for finished-output QA\n- `exact-svg-mascot` for protected named mascot identity\n- `share-community-demo` for explicit post-verification community publishing consent\n\nIf focused skill routing is not exposed, apply the same safety rule locally rather than claiming the skill was invoked.\n\n## Final report\n\nReturn:\n\n- selected execution path\n- observed capabilities used\n- side jobs actually executed\n- artifacts actually created\n- identity source/integrity summary for named logos/mascots\n- narrative shape and narrative-gate verdict\n- perception-preflight verdict\n- complete still taxonomy verdict\n- severity counts and cumulative visual-slop pressure\n- targeted repairs actually performed and their responsible dimensions\n- motion/render verdict when applicable\n- final verdict: `PASS`, `FAIL:fixable`, or `HOLD`\n\nNever claim a tool, agent, sandbox artifact, render, connected app, asset resolution/download, reference/demo inspection, or publication action that did not actually occur.\n"
}