Add v0.5.4 discoverability launch kit

[Cherwayway]

Discoverability Launch Iteration Contract

• Status: Day 0 implementation and verification; distribution not started
• Target Kit Version: 0.5.4
• Workspace Schema: unchanged at 1
• GitHub issue: #15

Feedback Signal and sanitized reproduction

The rolling GitHub Traffic window on 2026-07-29 reported 37 views from one
unique visitor and 83 clones from 36 unique cloners. The only referrer was
github.com; exact-name search found the repository, while broad intent
queries did not place it in the first 100 results. Current stars, forks, issue
activity, and known visitors could be traced to the team. The repository had no
public external backlink or platform-native Claude plugin discovery path.

The sanitized source snapshot and commands are recorded in
docs/research/2026-07-29-github-discoverability-diagnosis.md. Clone counts are
diagnostic only because CI, bots, IDEs, and Agent installs can clone without a
human repository visit.

Primary hypothesis and observable condition

Hypothesis: the earliest broken stage is acquisition, followed by an unclear
boundary with Claude Code Auto Memory and a high-friction install path. A
platform-native adapter, an observable fail-to-reuse demo, and three distinct
distribution paths should produce independently evidenced movement through
visit, activation, real evolution, and later reuse.

The seven-day condition is machine-readable in docs/launch/experiment.json:
30 GitHub-adjusted external unique visitors, three independently confirmed
valid activations, three real evolutions, and at least one later reuse record.

Smallest change and non-goals

The iteration adds only:

• one Claude Code marketplace adapter that delegates installation to the
  existing immutable-Release contract;
• one Auto Memory comparison in English and Chinese;
• one executed 74-second demo plus a short link-preview cut;
• one structured seven-day experiment and privacy-minimized snapshot/evidence
  tool; and
• channel-specific launch copy and landing paths.

Non-goals: no new runtime command, background process, telemetry, hosted
service, automatic global instruction edit, Workspace Schema change, or claim
that clone traffic represents people.

Boundaries and rollback

• Safety: the plugin is a discovery/install adapter only. It preserves
  immutable Release resolution, checksum and identity verification, dry-run,
  exact-plan approval, and the existing Bootstrap boundary.
• Privacy: evidence uses opaque participant IDs and boolean outcomes. It stores
  no raw conversation, complete log, contact detail, customer data, secret, or
  private absolute workspace path.
• Compatibility: Kit Version becomes 0.5.4; Workspace Schema remains 1. The
  runtime and prior stable update path are unchanged.
• Rollback: remove the marketplace/launch surfaces and stop distribution. No
  installed workspace data or schema migration needs reversal. A published
  Release remains immutable and would be superseded by a later version.

Verification

Completed on 2026-07-29:

• npm test: 203 tests, 183 passed, 20 expected platform skips, zero failed;
• deterministic plugin resolution smoke test plus official Claude marketplace
  and strict plugin validation: passed;
• executed demo generation: nine frames / 74 seconds plus four frames / eight
  seconds, with the expected failing test and repaired passing test enforced;
• offline evidence/snapshot outcomes and both interval boundaries: passed;
• authenticated Day 0 snapshot: one reported unique visitor, zero
  GitHub-adjusted external visitors, experiment not_started;
• npm run version:check: Kit Version 0.5.4 synchronized; and
• independent Standards and Spec reviews against fixed point a0bcec3: clean.

Release note

Kit Version 0.5.4 adds a Claude Code discovery adapter, executed terminal demo,
Auto Memory comparison, and a privacy-minimized seven-day acquisition and
activation experiment without changing Workspace Schema 1 or runtime write
authority.

External execution gates

Repository changes and Day 0 evidence can be completed locally. Marketplace
submission, public community posting, and direct invitations speak under a
maintainer's identity and remain pending until the maintainer explicitly
approves those external actions and target audiences. Their results must be
appended as new evidence records; historical records are never rewritten.