ideation-sandbox: codify external-precedent web_search step before Discussion authoring

Author's Note: Filed by Claude Opus 4.7 (Claude Code) during session b5a17132-7324-46e1-b73e-038825bb4d55 after @tobiu's challenge this turn surfaced a missing discipline. Empirical anchor: this very session-arc almost shipped a Neo-native ad-hoc payload schema for #10334 that diverged from the Google Agent2Agent (A2A) Protocol — an established Linux-Foundation v1.0 standard adopted by 150 production orgs (LangGraph/CrewAI/Google ADK/etc.). The catch came only when @tobiu suggested "a quick web_search exploration" before implementation. Without that prompt, the schema would have re-invented existing infrastructure at exactly the wrong moment. Codifying the discipline so future ideation flows fire it structurally.

Context

The Ideation Sandbox skill (#10278 + ideation-sandbox-workflow.md) governs how exploratory architectural concepts move from sketch → Discussion → Epic graduation. Current §1 has a content-level adjacency sweep (per #10276) that checks resources/content/issues/ and resources/content/discussions/ for in-flight similar work. That covers internal duplicate-detection.

It does NOT cover external-precedent verification — checking whether the structural pattern being proposed (protocol / schema / interface / state-machine / message envelope / cross-system convention) already exists as an industry/community standard.

The Problem

Ideation that proposes structural patterns with potential industry equivalents is at risk of re-inventing existing standards. Examples:

• Agent-to-agent communication protocols (A2A — actually exists per this turn's discovery)
• Memory/embedding schemas (LangChain/LlamaIndex conventions)
• State-machine task lifecycles (Temporal, GitHub Actions, BPMN)
• Tool-use protocols (MCP — already aligned, but the precedent-check would have surfaced it earlier in design)

When Discussion bodies reference "open X Protocol" or "industry standard for Y" without sourcing, two failure modes:

1. Vague-reference accepted: the protocol is real but unsourced → schema diverges from the standard at implementation time
2. Vague-reference fabricated: no such protocol exists → wasted "alignment" effort

Both prevented by a structural web_search step at ideation-authoring time.

Why industry-friction-radar Doesn't Fit (per @tobiu's invited challenge)

Reading industry-friction-radar/references/industry-friction-radar-workflow.md:

"When executing this radar, you are strictly forbidden from searching for or adopting 'SOTA' or 'Industry Standards.' You must target the Frontier — the bleeding edge where those mainstream standards are breaking down and failing. We do not ingest the standards; we ingest the friction points caused by those standards failing at scale."

Radar = frontier-friction-extraction (search for what's NOT working in standards; abstract the friction; build Neo-native solutions). The A2A finding is the OPPOSITE — an established mainstream standard radar is forbidden from surfacing. Distinct disciplines, opposite directions:

Both legitimate, complementary, non-overlapping. Radar would have refused to surface A2A; precedent-check would have mandated it.

The Architectural Reality

Neo is an Application Engine (per feedback_neo_is_engine_not_framework) — engines like Unreal/Godot adopt standards (USD, glTF, OpenXR) without losing identity. Speaking established protocols strengthens engine identity (interoperability) rather than dilutes it. The right discipline is verify-then-align-or-diverge-with-rationale, not avoid-all-standards.

The Fix

Add a new sub-step to ideation-sandbox-workflow.md §1 (or §2 Initial Proposal phase) — External-Precedent Sweep:

<h3 class="neo-h3" data-record-id="7">§1.X External-Precedent Sweep (per #ThisTicket)</h3>

When ideating about structural patterns with potential industry equivalents — protocols, 
schemas, interfaces, state machines, message envelopes, cross-system conventions, agent-
coordination patterns — you MUST execute a web_search to surface established standards 
BEFORE filing the Discussion.

**Trigger conditions** (fire if any apply):
- Ideation references "open X Protocol" / "industry standard for Y" / "established convention"
- Ideation proposes a SCHEMA / INTERFACE / PROTOCOL / STATE_MACHINE definition
- Ideation describes agent-to-agent / system-to-system / cross-vendor patterns
- Body uses words like "protocol", "spec", "standard", "convention" suggesting external precedent

**Skip conditions** (no web_search needed):
- Pure Neo-internal substrate (boot orientation, MX framing, hemisphere split, daemon scheduling)
- Codebase-specific debt or refactor
- Already-cited external precedent with verifiable URL

**Execution:**
1. Run web_search with current year + protocol-domain keywords (e.g., "agent-to-agent protocol standard 2026")
2. If a standard surfaces: cite the canonical URL inline in the Discussion body Rationale; explicitly choose Align / Diverge-with-rationale / Hybrid
3. If no standard surfaces: document the search ("I searched for [keywords] and found no canonical industry standard; proposing Neo-native design")
4. The search documentation becomes part of the Author's Note or a §Background section

**Distinction from `industry-friction-radar`:**
- Radar targets frontier-friction (forbidden from surfacing established standards)
- Precedent-sweep targets established standards (mandates surfacing them)
- Both legitimate, complementary, opposite directions

Acceptance Criteria

• .agent/skills/ideation-sandbox/references/ideation-sandbox-workflow.md updated with new sub-step (suggested location: §1 alongside the #10276 adjacency-sweep, OR §2 Initial Proposal pre-filing gate)
• Trigger conditions explicit (when fires) + skip conditions (when doesn't)
• Distinction from industry-friction-radar codified inline (so future agents don't conflate)
• Empirical anchor cited: this turn's A2A discovery as the load-bearing example of why the discipline matters
• Cross-reference added in industry-friction-radar workflow noting the complementary discipline (so radar invocations don't accidentally try to do precedent-check work)
• Optional: Discussion #10313 (A2A Task Schema) updated retroactively with the cited A2A standard URL (cleanup post-merge if author concurs)

Out of Scope

• Replacing industry-friction-radar — this is complementary, not replacement. Radar's frontier-friction discipline is distinct and remains valuable.
• Mandatory web_search for every ideation — not all Discussions touch external-precedent territory. Trigger conditions explicit so Neo-internal ideation skips correctly.
• Source-credibility scoring — first-pass surface-the-existence-of-standards is sufficient. Citation evaluation can layer on later.
• Live spec-import / SDK adoption — this is a precedent-AWARENESS step at ideation time. Implementation-time alignment decisions stay scope of the implementing PR/ticket.
• Anthropic-specific search domain restrictions — this is a generic web_search; per-domain filtering is configurable but not mandated.
• Backporting precedent-check to existing Discussions — except for #10313 cleanup (optional AC), no retroactive sweep.

Avoided Traps

• Conflating with industry-friction-radar — rejected; opposite scopes (frontier-friction vs established-standards). The challenge from @tobiu surfaced this trap explicitly.
• Making it a separate skill — rejected for v1; it's a single-step discipline, not a multi-phase workflow. If scope grows, can graduate to its own skill later.
• Hard-mandating web_search for every ideation — rejected; over-fires on Neo-internal ideation (e.g., #10309 boot orientation has no industry precedent). Trigger conditions narrow the surface.
• Auto-fail Discussion if precedent-sweep skipped — rejected; discipline is reviewer-flag-able, not mechanical-block. Authors can document "I searched, found nothing" or "skip-condition applies."
• Building a curated registry of external standards — rejected as v1 scope; web_search at ideation time gives current snapshot. Static registry would go stale. Could be added later if precedent-search becomes high-frequency.

Related

• ai/skills/ideation-sandbox/ — the skill being extended.
• ai/skills/industry-friction-radar/ — complementary skill; explicit cross-reference to be added.
• #10276 — pre-authoring adjacency sweep (precedent for "mandatory pre-filing step in ideation"); structural model for this addition.
• #10283 — pre-authoring meta-skill sweep for create-skill (parallel pattern for adjacent skill enhancements).
• Discussion #10313 — A2A Task Object Schema; the empirical anchor where this discipline would have caught the unsourced "open A2A Protocol" reference at ideation time instead of cycle-2 substrate review.
• #10334 — A2A-aligned schema work currently in flight; would have benefited from precedent-check at ideation time.
• feedback_verify_written_claims_against_precedent memory — durable anchor for the verify-precedent-before-implementing reflex.

Origin Session ID: b5a17132-7324-46e1-b73e-038825bb4d55 Retrieval Hint: "ideation-sandbox external-precedent web_search alignment standards-verification radar-distinction A2A discovery healing-amnesia organism"