Lintel on GitHub Copilot. Generated from
skills/frontend-shader/SKILL.md; edit the canonical file, then runli-copilot init.
- Resource root:
../../..from this skill's base directory (the Lintel source withbin/,lib/,skills/). Write plans, state and evidence into the working repository's.claude/tree, never into the resource root.- Skill-relative paths: this skill's own
scripts/,references/anddata/folders (and a<base>that the workflow defines as its own directory) mean../../../skills/frontend-shader/in the Lintel source, not this generated folder.${LINTEL_SKILLS_DIR:-skills}means the skills root,../../../skills. Abin/li-runstep runs in the working repository, so use$LINTEL_SKILLS_DIR/frontend-shader/there.- Shell steps: run Bash snippets with Bash (Git for Windows'
bash.exeon Windows, neverSystem32\bash.exe). Save a snippet to a temporary.shfile and runbash "<resource root>/bin/li-run" <file>; it preparesLINTEL_SOURCE_ROOT,LINTEL_REPO_ROOTand the profile context.- Tools: Read=
view, Write=create, Edit=edit, Bash=bash/powershell, Grep=grep, Glob=glob, AskUserQuestion=ask_user, TodoWrite=the plan checklist, Task or a named role=taskwith that custom agent, WebFetch=web_fetch.- Other Lintel workflows are native skills: invoke
/li-<name>rather than reading their files. Named roles such asCodeReviewerare custom agents.
You are the frontend-shader sub-skill — shader-engineer for the v3.7 frontend-* family (Phase A2).
What this skill does
Reads operator brief → ShaderEngineer chooses no shader or an evidenced,
project-compatible renderer, with visual thesis, source-backed snippet references,
GPU fallbacks and performance budget → writes shader.json (schema_version: 1).
Solo-invokable for component-mode or auto-invoked by the /li-frontend-design orchestrator in parallel-dispatch (Workflow Step 4) as the third parallel sub-skill (after typography + motion).
L-001-discipline: skill body is the contract. Agent at invocation produces specific library picks + GLSL recommendations. Don't pre-bake shader-snippets in the SKILL.md body.
Use the shader definition in the shared design contract.
Validate visual_thesis: none with library: null before GPU-only requirements.
When to use
- Solo: "hero-background for enterprise SaaS landing — want a subtle mesh-gradient"
- Orchestrator-parallel: dispatched from
/li-frontend-designStep 4 - Audit existing site: "extract shader-thesis from this site"
When NOT to use
- Static site, no hero-visual ambition → shader overkill
- 3D scene-graph implementation needed → use the project's supported scene tooling (this method chooses, not builds, scenes)
- CSS-gradient is enough → agent surfaces "no shader needed" and short-circuits
Inputs
- Required
--brief <text>(one minimum) - Optional
--visual-thesis <mesh-gradient|noise-field|fluid-sim|particle-system|displacement-warp|none>— default: inferred from brief - Optional
--perf-budget <low-end|mid-tier|high-end-only>— affects GPU-fallback-strategy - Optional
--out <path>— output path (default: stdout solo,$run_dir/shader.jsonorchestrator) - Optional
--customer-share— selects the customer-share control boundary; no automatic license validator
Workflow
Step 1 — Parse + warm context
brief="${BRIEF:-${1:-}}"
thesis="${VISUAL_THESIS:-auto}"
perf="${PERF_BUDGET:-mid-tier}"
out="${OUT:-}" # absent --out means stdout, not a path to validate
[ -z "$brief" ] && { echo "Need --brief"; exit 2; }
Step 2 — ShaderEngineer agent dispatch
Use agents/frontend/ShaderEngineer.md as the decision method and follow
axis ownership. Delegate only
when a separate context is useful and actually available; the role returns a
draft and this caller owns Step 3's single publication. First test whether static
artwork or a supported CSS treatment meets the brief: visual_thesis: none
and library: null are successful results. For an actual GPU need, compare
existing project tooling, required effect, framework/browser/device support,
measured or explicitly unverified cost, and fallback behavior. Choose a renderer
only on that evidence. A snippet/function library is not a renderer, and a
vendor name cannot establish device safety.
Retain exact selected-release/file terms and attribution for the renderer,
snippets, imports and artwork. Keep the existing lygia_imports field when
applicable; the field name does not require selecting that library.
Use actual supplied/verified license evidence or request an authorized lookup. A role name does not prove current terms were checked.
Step 3 — Produce shader.json
{
"schema_version": 1,
"generated_at": "<iso-8601>",
"brief_summary": "<one-line>",
"visual_thesis": "<mesh-gradient | noise-field | fluid-sim | particle-system | displacement-warp | none>",
"library": {
"name": "<selected project-compatible renderer>",
"version": "<exact selected release>",
"license": {"type": "<verified selected terms>", "source": "<primary release/file source>"},
"npm": "<selected package, if applicable>",
"operator_instruction": "<authorized project-local setup; no automatic install>"
},
"glsl_snippets": [
{
"name": "mesh-gradient-3color",
"purpose": "hero background",
"source": "<actual selected file/release reference>",
"notes": "<supported parameters and source/profile token mapping>"
}
],
"lygia_imports": [],
"perf_budget": {
"fps_target": 60,
"max_draw_calls": 4,
"fallback_strategy_low_end": "swap to CSS conic-gradient",
"fallback_strategy_no_webgl": "static CSS gradient + noise SVG",
"respect_prefers_reduced_motion": true,
"intersection_observer_pause": true
},
"gpu_thesis": {
"complexity": "low | medium | high",
"mobile_strategy": "downscale-resolution-50% | disable | full",
"explanation": "<selected device/browser/resolution and actual measured frame/GPU evidence, or explicitly unverified>"
},
"operator_instructions_md": "<setup for the actual selected renderer only; use verified profile/brief colors, real reduced-motion and no-WebGL fallbacks, and cleanup instructions; no automatic install>"
}
Agent fills in specific picks. Don't hardcode.
Step 4 — Schema-validate + emit
Keep Step 3's actual parsed JSON object as fragment until validation and any
required licensing checks finish. In the trusted source Python scope, repo is
the explicit target root and out is None when --out was omitted, otherwise
the literal repository-relative output path. For named output, the authorized
caller captures original_output_state through P03 before generation (None
means originally absent, not overwrite permission). Then execute:
import sys
from design_contract import emit_fragment
try:
emit_fragment(fragment, "shader", repo=repo, out=out,
original_output_state=None if out is None else original_output_state)
except (ValueError, OSError, UnicodeError) as error:
print(f"ERROR [lintel/design]: {error}", file=sys.stderr)
raise SystemExit(2)
This emits only the validated fragment, including the no-shader branch, not a
success-shaped receipt. Invalid data emits no stdout or named file; publication
errors have a nonzero exit. Never pass stdout/special/absolute paths to the rooted
reader. For --customer-share, use the customer-share control boundary
on the same data or an owned relative staging file. Missing mandatory licensing
or policy observations remain unverified; stdout does not exempt them.
Use its named shader-licensing procedure for selected renderers, snippets,
imports and assets. Required brief/profile obligations apply without this flag too.
Step 5 — Visual-thesis === "none" short-circuit
Agent can return visual_thesis="none" if the brief doesn't warrant a shader. Skill body STILL emits valid JSON so orchestrator-Step-5 synthesis can handle shader: null gracefully.
This branch is validated before active-shader library/performance checks, not after
a failing mandatory-library check. Emit no canvas, GPU import or install instruction.
For an active shader, retain the real fallback/reduced-motion budget and selected
release/source/license evidence. A CSS media query alone does not stop a JS GPU loop.
Reusable patterns
Follow the reusable pattern consumer contract.
Direct entry resolves, or verifies a supplied lock or projected pattern_context, before
choosing. Mandatory clauses bound the shader decision; defaults apply only where the brief did
not decide; unconstrained choices follow the usual Design DNA brief > profile > corpus rules.
Record the clause IDs each choice satisfies; prose clauses need ordinary evidence review.
Pattern text is not evidence of licensing or accessibility.
Status protocol
- DONE — shader.json written, schema valid, library + perf-budget non-empty (OR visual_thesis="none")
- DONE_WITH_CONCERNS — picked library has commercial-tier requirements (rare)
- BLOCKED — brief unparsable, OR customer-share license-check failed
- NEEDS_CONTEXT — brief lacks visual-direction (cant determine if shader needed)
Pause-points
- Customer-share + library has obscure license → flag explicit
- Brief mentions specific shader-lib agent doesn't know → may need NEEDS_CONTEXT
- visual_thesis="none" — short-circuit confirmation to operator (no shader is fine)
Integration
Reads:
--briefargument- Explicitly selected and authorized shader assets or verified profile references; no personal-home discovery or lazy-created asset folder
Writes:
shader.json(stdout default, $OUT-path if orchestrator)- Audit-log:
.claude/runtime/audit/frontend-shader-runs.jsonl
Calls into:
agents/frontend/ShaderEngineer.md(primary)- The
shader-licensingsource-inspection procedure in the customer-share control boundary;/li-compliance-gateevaluates its recorded outcomes, not shader licenses
Consumed by:
/li-frontend-designWorkflow Step 5 (synthesis input —shaderfield)- Operator direct (solo component-mode)
Anti-patterns
- Hardcoding "always Paper Shaders" — L-001 violation. Agent picks based on brief.
- Skipping perf_budget.respect_prefers_reduced_motion — accessibility-fail. Required field.
- No fallback_strategy_low_end — mobile-users will see broken page. Required.
- No fallback_strategy_no_webgl — WebGL-disabled browsers (rare but real) see broken page. Required.
- Producing shader.json without
schema_version— M-5 compliance.
Failure recovery
- Brief too vague → NEEDS_CONTEXT with question ("subtle mesh-gradient or full 3D scene?")
- Library-version-recommendation outdated → agent re-picks at invocation
- Schema validation fails: BLOCKED + diff
- visual_thesis="none" but operator wanted shader: surface "consider CSS-gradient instead"
Recommended next steps after invocation
- Solo: review shader.json + apply to target project
- Orchestrator: parallel-dispatch returns to
/li-frontend-designStep 5 - Customer-share: obtain the required
shader-licensingevidence, then evaluate the actual controls through/li-compliance-gate; unresolved permission blocks sharing - Future: propose a reviewed pattern at an explicitly selected authorized destination; no automatic personal write or activation follows from a shader decision