Scripts-First Restructure — rebuild scripts-first-workflow-architecture as JSON-routing-only + autonomous engine
18 sections
Overview
Rebuild scripts-first-workflow-architecture into a JSON-routing-only, flat-node, linear-route plan driven by ONE autonomous script-first engine — making it finally embody the thesis it defines (it never dogfooded its own JSON + scripts-first design). Stage spine (linear): fix the converter → self-convert the subject EARLY → author all further work natively in JSON → harden via dogfooding → migrate the whole corpus → strip every structural markdown parser.
Planned
3 sections
Strip skill markdown-walk (consuming skills -> v7-only)
SUPERSEDED per `decisions/05-three-plan-types.md` — the markdown-walk strip is CANCELLED. The corpus is a PERMANENT three-type system (json-static + json-dynamic + legacy-markdown); markdown plans + the markdown-walk path + the structural markdown parsers are RETAINED permanently, never stripped. §10 does NOT establish a 'no markdown plan remains' precondition (that precondition no longer exists). The original (now-cancelled) goal was: remove the markdown-walk path the §09 additive rewrites retained so the orchestrator skills become v7-only. Retained here as a frontmatter record of the superseded intent; the section ships no strip work.
Strip structural markdown parsers (now-dead provider)
Delete every STRUCTURAL markdown-parsing surface (read.py md branches, write.py legacy-body writers, content_hash.py md funcs, _frontmatter.py, plan_io.py md surface) AND the deprecated v6 checkbox scanner (scripts/plan_corpus/checkbox_scan.py) once the corpus is uniformly JSON-first AND §11A has removed every skill consumer of these parsers. Under v7 (§07A) status is work_items[].status — engine-written JSON read by serve_next, NOT a markdown checkbox scan — so the v6 status scanner is DELETED here, not retained. RETAIN only the JSON-native content surfaces (write.py:update_sidecar, render.py --view, read.py:get_sidecar). This is the PROVIDER strip and MUST land AFTER §11A (the consumer strip): the parsers are dead only once no skill markdown-walk path imports them. Gated on §10 corpus migration + §11A skills strip.
Verification + autopilot soak — SUPERSEDED (descoped at plan close-out, 2026-06-03)
SUPERSEDED — descoped by user decision at plan close-out (2026-06-03). NOT superseded by redundancy (unlike §11A/§11B, which Decision-05 made unnecessary): this is a deliberate descope of the verification capstone. The delivered engine (§02-§11C) ships WITHOUT the planned end-to-end proof — the aggregate per-node acceptance suite registration in compiler_repo/test-all.sh, the fake-clock deterministic soak, the real >=4h autopilot soak evidence artifact, and the live end-to-end /continue-roadmap --autopilot cross-plan run were NOT executed. Original goal (retained for record): prove the rebuilt system end-to-end via aggregate acceptance tests, cross-plan determinism, and the autopilot soak — fully autonomous cross-plan for hours, never halting (auto add-bug/fix-bug, jump-and-return), validating invariant 8 + the zero-halt engine. Open tooling bugs against this plan's deliverables remain tracked separately in bug-tracker/section-07-tooling-cli.md.
Completed
15 sections
Current-system evaluation — inventory, bug-to-justification map, umbrella-section dispositions, todo-YAH UX requirement
Produce the durable evaluation artifact justifying the restructure scope: full inventory of the current /continue-roadmap convolution (17 gates + 105 orchestrator symbols per `references/current-system-inventory.md:15` + 14 umbrella invariants + 54 umbrella content nodes — the umbrella plan has been converted to v7 JSON by this plan's own §04/§10 work, so `plans/completed/scripts-first-workflow-architecture/` now carries `content/*.md` sidecars, NOT top-level `section-*.md` files), per-bug justification map (which bugs forced which convolution), umbrella-section disposition table (SUPERSEDED / SUBSUMED / RETAINED per section, with replacement target), todo-YAH UX requirement (post-restructure visibility pattern). Output lands AS deliverable into `plans/completed/scripts-first-workflow-architecture/` via supersession markers + an evaluation appendix; this section owns the lift.
Principles & Invariants — north-star the rebuild derives from
Author the consolidated Principles & Invariants artifact (the 20 north-star
Corrected converter — per-plan plan.json (routing-only) + flat content sidecars
Ship a production converter that emits, per plan, a single readable plan.json carrying ROUTING ONLY {id, slug, key, status, hint_needs, body_ref} plus flat content/<id>--<slug>.md body files holding ALL content (items as id-keyed markdown checkboxes); reuse 006.build_plan_json_from_dir as the sole parser SSOT; emit PER-PLAN only (no dependency on the cross-plan merge engine).
Minimal JSON-authoring substrate — schema + write API + id-keyed scanner + per-plan serve_next
Land the minimal substrate that makes a converted per-plan plan.json executable WITHOUT the cross-plan merge engine: routing-only plan.json schema + atomic write API + net-new id-keyed checkbox scanner (fenced-block-skip) + per-plan serve_next built on the reused graph.py algorithm. Bootstrap prerequisite for §04 self-convert. Close-out verification gates (pytest substrate suite, negative-import test, plan_corpus check, /tpr-review final) migrated to §03A per deadlock-cure 2026-05-28 (Option 1).
Substrate verification gates (migrated §03.N close-out)
Run the substrate verification gates that previously gated §03 close-out (pytest
Self-convert the subject plan to JSON-first (the dogfood)
Convert scripts-first-workflow-architecture to JSON-first using the §02 converter
Per-plan SSOT + on-the-fly cross-plan route
Make per-plan plan.json the SSOT and derive cross-plan execution order on-the-fly
Autonomous-execution engine — one group, flat capability nodes
Collapse the 6 patched autopilot sections (§26/27/28/31/32/34/36 of the subject
commit-push redesign — path-scoped staging; structural gates removed
Make /commit-push stage ONLY the active node/plan''s path-set (parallel-work-friendly)
Route schema v7 — sections + work_items split (supersedes v6 flat-nodes)
Redesign the per-plan plan.json from v6 flat-nodes (nodes[]) to v7 two-level
Route-access enforcement — PreToolUse hard-ban on LLM plan.json access
Author a PreToolUse hook (modeled on block-spec-edits.sh) that hard-bans the
Consuming-skill rewrites — ADDITIVE v7 path alongside untouched markdown; dogfood
ADDITIVELY teach the consuming skills (/continue-roadmap, /commit-push, /fix-bug,
Structural Cross-Plan Dependency — schema field + merge-layer hard-defer + cross-plan acyclicity
Make cross-plan dependency a first-class STRUCTURAL routing-engine concept — a
Corpus migration + consumer re-point + DROP-default re-screen
Per `decisions/05-three-plan-types.md`, the corpus is a permanent THREE-TYPE system (json-static + json-dynamic + legacy-markdown), NOT a uniformly-v7 corpus. Convert the plan dirs (+ bug-tracker/plans) EXPLICITLY opted into JSON via the §02 converter; legacy markdown plans stay markdown unless explicitly converted (the markdown-walk path is RETAINED per Decision-05, never stripped). Re-point cross-repo consumers (intel_repo importer, website_repo loaders) to PREFER plan.json natively while KEEPING the markdown-fallback branch so legacy plans still parse. Re-screen the 59 philosophies / 17 decisions via DROP-default gated on a citation/dependency map.
JSON Dynamic plans — dynamic marker + grow-on-close engine step + /create-plan --dynamic
Land the json-dynamic plan type (INV-22 / decisions/05) — a v7 plan that starts as a