Roadmap
Live from the repository's committed releases ledger (releases.db) — the same source of truth the repo's own dashboard renders. Status markers: ✅ done · 🚧 in progress · ⏸ paused. Each entry links to its GitHub issue.
8 items are in progress. The current long-horizon focus is named in the repo's roadmap dashboard.
In progress 8 items
- 🚧GH-201 · Gen 3.5 follow-ups — remaining ATE hardening arc (tasks 3–8b of #174)#201
- 🚧GH-141 · make Fuzzing and ATE actually useful — Phases 1, 2, 4, 5 (+ #142 prerequisite); Phase 3 deferred pending #143's incidence comparison#141
- 🆕GH-5 · kernel robustness: node:test unit runner#5
- 🆕Vendor legacy SWE skills and make skills/ canonical#325
- 🆕ci: hermetic local gate runner (retire GH-564 clone requirement) and offline pre-push fallback#496
- 🆕Builder can approve and close its own relay; no merge path checks for a reviewer (evidence, severity, fixes)#505
- 🆕Relay terminal authorization: prove a review happened without trusting builder-writable state#509
- 🆕Reconciler provenance lifecycle#591
Queue / parked intake 52 items
- NoneGH-223 — pre-push gate push double-applies through ref lock#223
- NoneGH-215 — wave_reconcile.py and roadmap-dashboard.sh hardcode pre-vendoring path depth#215
- NoneGH-216 — marathon-plan.sh ledger parser rejects link-style ROADMAP.md bullets#216
- 🆕GH-75 · single-page HTML dashboard: releases (mid/long term) + roadmap (immediate) in one read-only view#75
- 🆕GH-61 · RELEASES ledger durability hardening (GH-57 follow-up)#61
- 🆕GH-67 · Commandcode builder default widened to `--yolo` — closer evaluation → possible build#67
- 🚧GH-32 · RELEASES app: SQLite-backed release ledger, CLI-only writes, generated RELEASES.md, cross-repo UI#32
- 🆕GH-28 · RELEASES.md ledger discipline: deterministic bloat checks + per-release tracking issue#28
- 🆕GH-17 · SOP for evaluating new agent harnesses and frontier models#17
- 🆕GH-18 · Harness evaluation: Command Code (cmd) and model matrix#18
- 🆕marathon-drive blocked-before-dispatch refusal omits XYZ_ARCHIVE_ROOT#255
- 🆕No preflight checks the builder can reach a phase artifact paths#256
- 🆕Jog: serial immediate-queue execution engine and skill (Phase 1)#259
- 🆕Medium-level write-ops logging of agent disk-write commands#275
- 🆕export_timeline: draft release cards render before active cards — open-bucket sort ignores status#254
- 🆕feat(ate): Gen 4 ATE & True Evolutionary Fuzzing — Feedback-guided mutational engine, semantic domain oracles, and adaptive pairwise matrix#299
- 🆕Speed up harness -> gateway -> model resolution with a real cache#346
- 🆕Sleep-vs-readiness audit: one confirmed race in the agent-chorus doorbell assertion, and the case against the paid Test Reliability tier#345
- 🆕CI: attest Linux as an unprivileged user — the canary's 14 failures are almost all EUID=0 artifacts#341
- 🆕Explore a secure cross-device AgentChorus bridge over Cloudflare Tunnel#384
- 🆕Harness root resolution: one resolver, two roots, pinned — retire RADAR-class-vendored-root-resolution#396
- 🆕Let the profile carry the route#399
- 🆕Board sync: auto-add issues to the Projects board when any agent starts work#402
- 🆕External review (Russ K.): eight findings on guards that cover a narrower path than the doc claims#406
- 🆕four documentation trues-ups#416
- 🆕Marathon planner still reads the frozen ROADMAP.md#418
- 🆕Wave reconciliation is never invoked automatically — wire it post-merge, but fix its ledger write first#421
- 🆕releases roadmap render: emit the DB as ledger markdown — the missing verb #418 actually depends on#423
- 🆕roadmap_items.status_marker has no CLI writer — in releases-mode a row can never leave 🆕#424
- 🆕--gate's provenance check never compares the PR number — it proves TESTS-RESULTS/ is non-empty#425
- 🆕marathon-triage: verify PRS ratings, then compute, preflight, and dry-run the marathon before it fires#443
- 🆕runtime:parity — turn shims reject --help when RELAY_AGENT is unset under XYZ_PYTHON=0#445
- 🆕marathon_plan.py writes unprompted plan file to disk on default invocation without --dry-run#446
- 🆕Start publishing release tags (git tag + GitHub Release per shipped ledger release)#452
- 🆕wave_reconcile dies on an unnamed release, and enforces PDDA full-mode on an observe-mode repo#454
- 🆕ATE/Fuzz campaign vs the model-alias resolver — counterexamples must land as fixes#460
- 🆕Governance document identity audit#453
- 🆕Deploy Skills#484
- 🆕consult.sh: agy lane answered from a months-old ~/.gemini scratch copy, not the throwaway worktree#481
- 🆕relay-drive.sh --artifact-file silently seeds nothing — the reviewer never sees the artifact#482
- 🆕roadmap move/update --section help advertises section names the renderer drops, and nothing validates the value#491
- 🆕Nothing reconciles roadmap_items against issues closed outside a merged PR#492
- 🆕ATE exit-cleanup / runaway-process guard — gen4 adaptive-ate test hung ~3 days at ~100% CPU#478
- 🆕Migration 007: updated_at across all releases.db tables, so consumers can detect what changed#355
- 🆕skills: new co-author skill "keel" — balance modularity, flexibility, maintainability, and ponytail for spec writing and total refactors#514
- 🆕jog: operator-confirmed merge path ignores gh pr merge failure and records the task as completed#510
- 🆕test/agent-chorus.sh: telemetry assertions time-bomb on the default-ON pilot window#515
- 🆕Flight dashboard HTML design review#494
- 🆕harness: wire Meta's native muse CLI (Muse Spark 1.3 Contributor) as an XYZ route — catalog aliases, muse-turn.sh shim, four-surface evidence#518
- 🆕skills: new skill /five — the 5-and-5 decision checksum (5 key highlights + 5 explicit non-goals)#537
- 🆕feat(ledger): end-to-end removal of ROADMAP-DASHBOARD.md and view-staleness machinery#567
- 🆕fix(relay): honor Qwen/DeepSeek reasoning effort and correct false idle timeout attribution#608
Deferred · vision 1 item
Completed 111 items
Show 111 completed items
- ✅GH-228 · Org rename HiQS-Suite → HiQS-Labs: a hardcoded regex silently drops issue URLs from the roadmap shadow#228
- 🚧GH-232 · wave_reconcile should honor linked issue open/closed state before promoting multi-phase docs#232
- 🆕GH-169 items 3-4: repoint agent docs + dashboard-staleness push guard#243
- NoneGH-222 — releases update cannot re-point a release's tracking issue#222
- ✅GH-226 · xyz-vendor.sh transcript gate refuses repos that gitignore transcripts#226
- 🚧GH-233 · AgentChorus Gen 2 Phase 2 — Lifecycle Verbs, Watch Invalidation, Concurrency Suite & Citation Linter#233
- ✅GH-114 · headless agy -p stalls on TTY allocation / network waits until the idle watchdog kills it (exit 7)#114
- NoneGH-221 — agy-turn.sh auth pre-flight breaks on agy CLI >=1.1.19 — whoami subcommand removed#221
- 🚧GH-204 · BSD `sed -i ''` no-ops on Linux at production call sites#204
- ✅GH-113 · headless agy builder writes root scratch files, tripping containment (exit 6)#113
- 🚧GH-205 · validate.sh mutates four tracked files per run — gate not idempotent#205
- ✅GH-91 · a build turn has nowhere to write verification output#91
- 🆕GH-105 · vendor the RELEASES DB system + HTML timeline generator into the `.xyz` payload (optional add-on)#105
- ✅GH-168 · wave_reconcile.py hard-fails and rolls back on pre-existing drift unrelated to the reconciled PR#168
- 🚧GH-182 · self_healer --mode heal is a facade (containment refuses any real target) plus unsafe gate design#182
- ✅GH-50 · sandboxed git --track / branch -D half-applies and loses uncommitted work#50
- ✅GH-2 · test-suite run relocated an untracked file into .tick/orphan-backups/#2
- ✅GH-202 · wave_reconcile aborts on marathon-plan exit 5 (items held) and promotes capture docs for OPEN issues#202
- ✅GH-197 · two-tier xyz-vendor.sh: Tier 1 core-harness default, Tier 2 opt-in RELEASES overlay + onboarding SOP (GH-105 follow-up)#197
- 🚧GH-193 · AgentChorus Gen 2#193
- ✅GH-1 · suite-wide fixture containment + clone-identity invariant gate#1
- ✅GH-174 · Harness & Models Registry SQLite Migration: Per-Device Config, Reasoning Effort Tracking, AI Grading Hooks & Blog Generator#174
- ✅GH-115 · marathon-drive prematurely escalates productive multi-round reviews at the fixed round cap (exit 4)#115
- 🆕relay-xyz SKILL.md: explicit QA/Consult prompt template for headless agents (hotfix)#246
- ✅GH-8 · kernel boundary hardening — CLI numeric validation, task/agent format contract#8
- 🆕ubuntu canary: EUID=0 defeats chmod-based assertions#249
- ✅GH-180 · repro_builder crashes on timeout telemetry records (exit_code: null → TypeError)#180
- 🆕validate.sh reports python:test_python_layer.py as FAILED when pytest is merely absent#251
- ✅GH-181 · repro_builder emits non-reproducing reproducers from real telemetry (mis-tokenized unquoted command, rc 127 vs expected 2)#181
- ✅GH-183 · active_explorer env-family fuzzing unsound (base_env={} hardcoded, one always-deferring vector, ambient-env leakage)#183
- ✅GH-184 · committed scratch artifact `.relay-scratch/probe_telemetry.json` makes every real turn a tracked-file mutation#184
- 🆕wave_reconcile.py: linked-issue extraction is mention-greedy; rollback leaves regenerated artifacts behind#271
- 🆕/express: hotfix fast lane — fix + suite + ledger + born-complete docs + gateless development landing + reconcile#267
- ✅GH-195 · marathon-root-audit's blind spot: a direct `python3 marathon_drive.py` call#195
- 🆕releases-merge-resolve.sh: regenerate and stage derived view artifacts after the DB rebuild#272
- 🆕Full switchover to Releases DB — retire ROADMAP.md#269
- ✅GH-10 · prevent-half of containment: require_fixture adoption across the fixture-creating suites + adoption guard + ci-local identity bracket#10
- 🆕marathon-root-audit.sh is blind to python3-spelled driver invocations (GH-195 class)#273
- 🆕Recalibrate Jog as a serial supervisor over Marathon execution#280
- 🆕Connect /10days, /radar, and PARKED to the RELEASES DB (read-only seams)#107
- ✅GH-35 · 3-tier test suite selection (docs / utility subsystems / core) + CPU governance#35
- 🆕wave_reconcile archive line mangles separator-less ROADMAP entries#314
- ✅GH-42 · relay automation: supported Commandcode turn-taker#42
- ✅GH-170 · Agent2Agent: close transcript glitches and harden publishing#170
- ✅GH-165 · Post-Merge Wave & Marathon Lifecycle Reconciler (Docs, ROADMAP, DB, Views, and Planning)#165
- 🆕wave_reconcile resolves its five harness tools repo-root-relative — inert on every vendored install#358
- ✅GH-124 · eliminate end-of-day closeout friction — automated phase QA receipts, auto-PR creation, and clone lifecycle hygiene#124
- 🆕CI: move advisory Ubuntu canary off the pull-request path#347
- ✅GH-77 · `/standup` — session-scoped triage: what did I leave open, what is rotting, is the plan still right?#77
- 🆕Test-suite recalibration: finish tiering, prove parallel equivalence, and define smoke/promotion lanes#365
- ✅GH-153 · RELEASES dashboard sidebar + full-cycle rollup (technical spike)#153
- ✅GH-148 · DeepSeek Harness (dsh) integration & deepseek-turn shim for OpenRouter DeepSeek V4 Pro#148
- ✅GH-144 · Agent2Agent 3+ participant onboarding + read-only status quick wins#144
- 🆕audit and prompt for target ROUTER.md ROADMAP.md frozen status during vendored updates#353
- ✅GH-132 · feat(skills): formal /review-xyz code review skill & multi-model harness#132
- 🆕releases check receipt chain failure phrasing and scoped --rebuild#360
- 🚧GH-155 · 3rd Gen ATE & Fuzzing#155
- ✅GH-135..140 · Wave-1 follow-ups: consult preflight verdict, attempts-gate root, suite registration, twin-divergence record, SIGPIPE sweep+guard, utcnow swap#135
- 🆕Local debugging mock harness for GitHub Projects V2 API#405
- ✅#129/#130/#131 · Wave 1 of the Harness Driver & Relay Seam Hardening marathon — three preflighted driver/relay fixes#None
- ✅GH-111 · retire manifest FREEZE; tasks and marathons are DIALED IN to exactly one release, as a database state#111
- 🆕relay-block validator never runs on the driven path#410
- ✅GH-108 · pri/sev/appeal/effort — the canonical task rating system (calc sum + operator override)#108
- 🆕transient claim collision indistinguishable from durable loss#412
- ✅Execution checklist for GH-111 + GH-108#None
- 🆕launch-artifact marker authorises destructive rebuild#413
- ✅GH-101 · Feasibility Study: Promoting Programmatic Script Runner (`script_runner.py`) into Core Relay & Consult Runtimes#101
- 🆕tick log exempt from the foreign-cwd guard#411
- ✅GH-103 · technical spike: RELEASES SQLite → timeline-ui ledger viewer (RELEASES dashboard view)#103
- 🆕nothing checks whether an inline code comment is still true#414
- ✅GH-102 · Unify Telemetry Schema & Inspection Tooling Across Fuzzing and ATE#102
- 🆕guard hook covers 6 of 12 Tier-A entrypoints#415
- ✅GH-94 · research: programmatic tool calling & code-mode execution for harnesses, telemetry, and containment#94
- ✅GH-57 · test(releases): SQLite ledger fuzzing recipes & multi-scenario resilience suite#57
- ✅GH-45 · validate.sh must refuse to run from a linked worktree — an observed run corrupted the parent clone#45
- 🆕make the three marathon process rules deterministic#419
- ✅GH-39 · RELEASES app: one-way GitHub Project release-card projection#39
- ✅GH-23 · Kernel invariant: enforce path-overlap rejection on direct tick claim and tick scope#23
- ✅GH-4 · the pre-push gate does not travel with clones: fresh clones push unverified#4
- ✅GH-14 · appendEvent writes non-atomically, so concurrent readers can observe torn event files#14
- ✅GH-15 · parallel runs are unreliable in a fresh clone; the GH-528 contention retry is not honoring its contract#15
- 🆕releases ledger roadmap layer never generalised to a vendored install#349
- ✅GH-3 · improve-loop.sh --state-dir durability — provenance evidence must not evaporate#3
- 🆕`manifest unship`, the retraction verb#351
- 🆕feat(skill): /merge-cleanup — Worktree & clone consolidation, PR sequencing, and safe teardown#436
- 🆕start-task governed workflow skill#439
- 🆕Isolate launch destination checks from dirty caller source#447
- 🆕Homepage quality proposition#461
- 🆕Other Apps & Tools directory#465
- 🆕Consume Model-catalog v1 (Phase 1): vendored catalog + generated alias YAML + drift check + version telemetry#450
- 🆕feat(skill): add a preservation gate to workhorse for irreversible operations#471
- 🆕feat(skill): add /unstuck goal-movement interrupt#473
- 🆕Remove rendered MD views (ROADMAP-DASHBOARD.md & LEADERBOARD.md) to eliminate staleness guards#474
- 🆕ci: make new-branch skill/test pushes use bounded gates and clarify draft-review bypass#487
- 🆕Git Sync Pulse as a portable Skills Army HQ projection#508
- 🆕skills: express hotfix lane v2 — true direct-push mode, commit-driven reconciliation, recovery subcommand, and central telemetry#516
- ✅relocate: Deployed Skills moves INTO the live Git Pulse Sync checkout (supersedes #508 verdict)#536
- ✅merge-cleanup: primary on-disk checkout is not reviewed first; PRs merge into a tree that cannot receive them#523
- ✅/unstuck — queue simultaneous blockers, tripwire recurring blocker classes#528
- ✅agent-chorus: transcripts do not record the lab, model, or effort level behind a seat#524
- ✅Work-state event stream + concurrent connectors: GitHub Kanban first, HQ and Flightdeck next#549
- ✅merge-cleanup: promises merge + conflict-resolution + teardown, implements one and a half — six verified failure modes#534
- ✅Relay scaffold verdict vocabulary contradicts validate-relay-block#533
- ✅Relay attestation rejects scaffold-required marker placement#529
- ✅work backfill from ledger state + a review_ready producer in reconcile (GH-549 follow-up)#564
- ✅merge-cleanup: gate-run disposable clones preserved forever (telemetry-only .tick/ unverifiable)#561
- ✅feat(skills): add ci-optimize transferable CI/CD audit & optimization skill#578
- ✅add /sop and /whack-a-mole skills; route skill management to skills-army-hq; XYZ_SKILLS_ROOT#586
- ✅tick mutating verbs silently accept unknown flags#554
- ✅feat(ledger): end-to-end retirement of RELEASES.md in favor of releases.db#568
- ✅merge-cleanup B1 leaves releases.db.bak after resolution#555