CHANGELOG.md

Summary of changes from v3.5

What	Change
Version	3.5 → 3.5.1
Repo structure	Modularized playbook into docs/playbook/ (12 section files). SEED-Method.md becomes kernel + index with stubs.
Agent context	Root AGENTS.md rewritten as a compact, agent-agnostic kernel (~2.5 KB always-loaded). Added CLAUDE.md and GEMINI.md shims.
Distribution builds	Added dist/AGENTS.kernel.md (token-efficient, ~7.5 KB) and dist/AGENTS.md (single-file bundle, ~56 KB) with generator + check scripts.
skills.sh skill	Added skills/seed/ (SKILL.md entrypoint + PLAYBOOK.md) with generator + check scripts. Install: npx skills add jveres/The-SEED-Method@seed.
README	Expanded with Quick Start (kernel-first Option A + bundle Option B), per-agent setup (pi, Codex, Claude Code, Gemini, OpenCode), skill install, and "What's in this repo" table.
Guardrails	Added dogfood guardrail: repo's own bootstrap must always reflect latest playbook.
Decisions	Recorded: generated files committed (overrides guardrail 7) with freshness enforced by check scripts.
Cookbook	Removed stray YAML frontmatter that could confuse skills CLI.

---

Summary of changes from v3.4

What	Change
Version	3.4 → 3.5
§5 Snapshot sections	Reordered into two clusters: dashboard (Goal, Status, Next actions, Open questions) and detail (Plan, Tasks, Proof, Decisions, Lessons, Practices, Working set, Coordination). Dashboard cluster appears first so the top of the file answers "where are we / what's next / what's blocked" in one screenful.
"How to use this doc"	Owner status-check instruction updated: references the dashboard cluster (first four sections) instead of listing scattered section names
Owner responsibilities	"snapshot sections" → "dashboard sections (top of file)"
§3.1 DoD	Snapshot update bullet reorganized to name dashboard and detail clusters
§7 State clarity	Audit questions reordered to match dashboard-first layout (Status → Next actions → Open questions → In progress → Owner action)
§10a Reset table	Row order matches new snapshot order (dashboard sections first)
Appendix D	WORKLOG.md skeleton reordered to match; HTML comments mark cluster boundaries

Cumulative changes from v1.1

Version	Change	Net effect
1.2	Removed Micro mode	−1 mode, −1 decision point, −~80 lines
1.3	Unified BATON.md + JOURNAL.md → WORKLOG.md	−2 artifacts, −1 phase, −1 global rule
1.4	Added ## Lessons	+1 section, +1 DoD line
1.5	Replaced scoring with owner review guide	−numeric scoring, +concrete review steps
1.6	Removed orphaned plan/tasks skeletons	−2 appendices
1.7	Added idle housekeeping directive	+1 global rule, +1 section
1.8	Renamed Risks → Open questions; added scratch docs; added no-scoring prohibition	+1 renamed section, +1 section (§5b), +1 prohibition line
1.9	Removed obsolete SEED format migration	−1 subsection
2.0	Named the playbook: Single Owner, Leveraged Orchestration; clarified owner/agent framing	Identity established
2.1	Added state clarity check to audit protocol	+5 clarity questions in §7
2.2	Added guardrails system: defaults, owner-imposed, learned; promotion flow; audit/DoD/DoRD integration	+1 global rule, +1 section (§11), guardrail checks woven into §3.1, §3.2, §5a, §6, §7, §10
2.3	Review-driven clarifications: AGENTS.md identity, multi-agent worklog carve-out, approval format, meaningful-change examples, proof-cannot-run protocol, bootstrap proof exception, canonical-source guardrail reword, idle edit categories, handoff note template, owner responsibilities, orientation header	11 targeted fixes, 0 structural changes
2.4	Principles section, Owner dashboard pointer, integration-baseline wording, proof recording formats (local/CI/manual), proof-cannot-run reconciled with guardrail #2, one-time exception recording, external-tool thin-index rule, coordinator timing, agent-framework scaffolding clarification	+1 section (Principles), 12 targeted fixes, 0 structural changes
2.5	Edge-case consistency: Principle #4 "or referenced from it", runtime-stop + proof-cannot-run multi-agent carve-outs, multi-agent chunk definition, snapshot proof bullet aligned with canonical-pointer policy, exception-vs-amendment boundary, external-tool discoverability in WORKLOG	7 targeted fixes, 0 structural changes
2.6	Self-consistency: guardrail #8 exempts mandated artifact names, guardrail #6 allows WORKLOG_ARCHIVE.md / SEED_REQUEST.md, runtime-stop escalation uses §9.5 ladder, handoff notes declared transient with integration as durable record	4 targeted fixes, 0 structural changes
2.7	Release lifecycle: §10a post-release closeout (seal + archive + reset), release records as permanent artifacts in docs/releases/, WORKLOG_ARCHIVE.md skeleton with releases index, SemVer default, Decisions reset policy, multi-agent closeout ownership	+1 section (§10a), +2 appendices, 3 guardrail updates, 0 structural changes to existing phases
2.8	SEED lifecycle: SEED describes project not release, persists across releases, optional consolidation with audit trail in release record	+1 subsection in §10a, 2 skeleton updates, 0 structural changes
3.0	Identity: renamed from "SOLO playbook" to "The SEED Method"; added ## Participants to SEED skeleton; added Global rule #8 (Identity) — agent names session-scoped, owner names and role labels are stable identifiers; log/handoff templates include role + session name	Identity reestablished, +1 global rule, participant traceability model
3.1	Naming contexts: role labels in procedural/authority contexts, participant names in operational artifacts; checkpoint messages use names; log templates model name usage	+1 paragraph in rule #8, 3 template updates, 0 structural changes
3.2	Agent platform metadata: ## Participants gains Platform column (orchestrator · model); log entries / handoff notes capture actual session config; Global rule #8 expanded; cookbook #17 rewritten	+1 table column, +1 paragraph in rule #8, 4 template updates, 1 cookbook rewrite
3.3	Best practices system: ## Practices section (append-only, parallel to Lessons); automatic collection via DoD; practice → skill promotion path (§11a); Principle #8 "Learn both ways"; audit checks learning system health; cookbook #18	+1 principle, +1 section (§11a), +1 WORKLOG section, +1 DoD bullet, +1 DoRD item, +1 audit check, 1 cookbook entry
3.4	Field-tested refinements: WORKLOG proof is pointer-only to README (eliminates dual-doc drift); lightweight log format for non-behavioral chunks; audit allows immediate housekeeping-safe fixes; guardrail #9 (derived data computed, not hand-maintained); high-velocity cookbook recipe	+1 default guardrail, +1 log format, +1 cookbook recipe, 3 tightened rules, 0 structural changes
3.5	Dashboard-first WORKLOG layout: snapshot sections reordered into dashboard cluster (Goal, Status, Next actions, Open questions) and detail cluster (Plan, Tasks, Proof, Decisions, Lessons, Practices, Working set, Coordination). First screenful answers "where are we / what's next / what's blocked."	0 new sections, 0 new rules — pure layout optimization driven by field experience

Required repo artifacts: SEED.md, README.md, WORKLOG.md (+ WORKLOG_ARCHIVE.md when log grows, + SEED_REQUEST.md when needed, + docs/*.md scratch docs as needed, + docs/releases/release-<id>.md after each release).