Audit and prioritize package.json discovery keywords

Parent epic: #12225

Context

PR #12239 corrected the most obvious discovery-surface drift in package.json: the description now uses the concise organism apex, and several framework-trap / stale keywords were removed. The remaining keyword list still needs a separate pass. The operator explicitly called out that keyword ordering must be priority-based and that the team must genuinely challenge what belongs in the list at all.

V-B-A before filing:

• origin/dev:package.json currently has 88 keywords after PR #12239.
• The list mixes apex/Brain terms (agent-operating-system, active-hybrid-graphrag, dreamservice, professional-ai-engineering-team) with Body/application-engine terms (multi-threaded, multi-window, web-engine), generic JS terms (javascript, json, css), and product-example or low-signal terms (browser-email-client, web-based-email-client, monaco-editor, smooth-ui).
• Duplicate sweep found historical archived keyword tickets (#6462, #6830, #8357) and active parent #12225 / completed #12227 context, but no active ticket dedicated to priority ordering plus keep/drop challenge.
• Live GitHub search for open issues matching package.json keywords returned no duplicates.

The Problem

package.json keywords are a machine-facing and human-facing discovery surface. They are not the same as README prose, but they still encode Neo's identity. Leaving the current 88-keyword array in inherited order risks diluting the organism / Agent OS apex with generic or stale terms. It can also make the metadata read like SEO stuffing instead of a deliberate ranked signal.

This ticket must avoid an unsupported claim that npm keyword order directly determines ranking. The verified problem is narrower and enough: order and inclusion are consumed by humans, package registries, crawlers, agents, and downstream tooling as metadata. A high-noise list communicates poor priority even if a given registry treats the array as unordered.

The Architectural Reality

package.json is an ADR 0018 identity surface. It carries both:

• FACT-adjacent package metadata (name, version, repository, scripts), which must remain mechanically correct.
• FRAMING metadata (description, keywords), which must stay audience-segmented but compatible with the canonical apex from #12225 / ADR 0018.

PR #12239 intentionally performed a trap purge and first-pass recategorization; it did not perform a full priority taxonomy or inclusion audit. This follow-up owns that second pass.

The Fix

Audit and rewrite package.json keywords[] using an explicit priority taxonomy:

1. Define the intended consumer model for the keyword surface: package registry search, machine/agent discovery, and cold human metadata scanning.
2. Classify every current keyword as keep, drop, or defer, with one-line rationale.
3. Reorder kept keywords by priority, not by append history or alphabetic convenience.
4. Cap the list or explicitly justify retaining a larger count. High-value specificity beats broad stuffing.
5. Keep the top cluster aligned with the current identity apex and concise machine register: organism / Agent OS / cross-model team / Neural Link / Active Hybrid GraphRAG / DreamService / self-healing / Body engine terms as appropriate.
6. Challenge low-signal terms aggressively, especially app-example terms and generic tech terms that do not help the intended consumer understand what Neo is.

Contract Ledger Matrix

Decision Record Impact

Aligned-with ADR 0018. No ADR change expected. If implementation discovers a new rule for keyword count/order semantics that should govern future identity updates, file a follow-up against the neo-identity-update skill rather than burying it in this ticket.

Acceptance Criteria

• package.json remains valid JSON and the keywords array is intentionally ordered by a documented priority taxonomy.
• The PR body includes a keep/drop/defer audit for every current keyword or an equivalent generated table.
• The top keyword cluster reflects the Neo identity apex and machine-register framing, not the old web-framework/app-example center of gravity.
• Low-signal, stale, generic, or misleading keywords are removed unless their retention is explicitly defended.
• No forbidden framework-trap terms are reintroduced, including residual *-framework marketing terms.
• If GitHub repo topics or other external metadata should differ from package.json, the PR body documents the exact proposed out-of-tree update or explains the intentional divergence.
• Cross-family review is completed before merge per ADR 0018 identity-surface discipline.

Out of Scope

• Rewriting the already-updated package.json description from PR #12239.
• npm publication or registry-side experiments.
• GitHub topic changes performed by an agent; proposals are allowed, operator application is separate.
• Editing README, benefits docs, portal copy, or generated SEO output unless the keyword audit surfaces a narrowly scoped follow-up.

Avoided Traps

• Keeping all 88 keywords because they already exist.
• Alphabetizing the array and calling that priority.
• SEO stuffing with generic terms that bury the apex.
• Deleting all Body/application-engine terms and making the package undiscoverable to engine/runtime users.
• Assuming npm ranking semantics without evidence.

Related

• Parent: #12225
• Follow-up to: #12227 / PR #12239
• ADR 0018: Neo Identity Source-of-Truth Model

Origin Session ID: 87203e0b-c9ff-4a82-a67b-9a0dda760c32 Retrieval Hint: "package.json keywords priority ordering keep drop defer identity apex discovery metadata PR 12239"