Milestones
List view
Non-release-blocking research and contingent experiments for Codex output style, DeepSeek Harness, in-app Browser tooling, GitHub Copilot tiers and CLI, Microsoft enterprise hosts, portable MCP architecture, and Copilot Studio. Findings may create later implementation milestones; this work never displaces the 1.0 release. #96 owns the Copilot Studio adoption decision. #163 is its contingent implementation follow-up and remains here until #96 selects `adopt` or `experiment`; a `reject` decision closes #163 as not planned.
No due date•0/10 issues closedBenchmark orchestration only after 1.0 termination and 1.1 premium roles are stable. Measure Sol/Fable planning with bounded Luna workers, Spark, context/compaction, spawn and handoff overhead, plugin payload, quality, latency, quota, and direct-driver controls before recommending any worker lane.
No due date•0/3 issues closedPost-trust-contract maintenance: release docs/demos, document consolidation, harness rename, and controlled contributor-history cleanup. These tasks never displace delivery correctness, premium lanes, orchestration measurement, or CI evidence.
No due date•0/4 issues closedPost-1.0 CI/CD evidence and release enforcement. Audit independent CI value, prove fast-first/full-final behavior, detect release drift, and reconcile upstream versions through v1.98.0 without duplicating exact-tree proof or independent review.
No due date•0/6 issues closedSubmit the already-published, Windows-verified Codex SDLC Wizard release to the universal Plugins Directory and complete any portal-owned launch work.
No due date•0/2 issues closedPost-1.0 premium host lanes and harness hardening. Order starts with memory/source-of-truth audit, premium host roles, High-only planning/review, preserve-known-good updates, maintenance-command classification, PowerShell and skill metadata correctness, doctor and instruction ownership, lifecycle and ordered-goal support, privacy-safe feedback, consumer soak, and bounded review research. Contributor-ready bounded follow-ups include failed-proof summaries (#124), reviewer confidence/severity clarity (#131), one-step updates (#147), Windows after-hours command UX (#162), and dual-review progress (#164). Speculative Copilot Studio implementation does not block 1.1; #163 remains with #96 in the Research milestone until the adoption decision is `adopt` or `experiment`.
No due date•0/26 issues closedMajor trust-contract release for bounded autonomous delivery. Release-blocking order: 1. #178 — replace lexical model-policy attestation with a deterministic skipped-surface contract; fresh maintainer-authorized replan required. 2. #174 — complete truthful model-policy migration only after #178 resolves the exhausted regex approach. 3. #86 — make Sol High the only autonomous Sol effort and remove the xhigh maintainer exception; resume after #174. 4. #109 — complete the measured bounded-correction pilot on a newly authorized checkpoint after #86. 5. #141 — block executable Git configuration at delivery. 6. #127 — restore one canonical repo-local $sdlc workflow entry. 7. #92 — eliminate nondeterministic Windows proof cleanup. 8. #93 — keep Windows review evidence honest when PowerShell cannot launch. 9. #79 — run the exact frozen candidate through real Windows Desktop and CLI acceptance, including observed config/picker precedence before update, after update, and after restart. 10. #88 — publish and verify the immutable 1.0.0 artifact across GitHub, npm, Codex Desktop, Codex CLI, and plugin payloads. Completed foundations include #157 truthful partial-review classification, #158 forbidden-tool rejection, #167 a verified tool-free Fable lane, #172 exact Fable verdict-literal prompting, and #175 immediate final tool-free Fable review prompting with strict invalid-output rejection. #158 one-time meta-review exception is consumed and must not be reused or broadened. #174 exhausted its two bounded correction rounds with canonical proof green but a final reviewer-verdict receipt NOT CERTIFIED. Its candidates are frozen evidence only; #178 is the explicit decomposition, not a hidden third correction. Every remaining item maps directly to a 1.0 release objective. Proof, independent review, required CI, real-Windows acceptance, and public-artifact verification must bind the intended candidate. No due date is claimed because the final two gates require maintainer hardware and publishing credentials.
No due date•16/26 issues closed