Design
What the CLI is, and how it is specified.
| Document | Status |
|---|---|
local-interface-verification-2026-09-13.md | WI-100 verification. Automated contracts, platform/consumer evidence, cost measurements and operator task walkthrough; no independent adoption claim |
local-interface.md | Implemented 2026-09-13 in WI-100. Foreground interface command, snapshot/read states, process and HTTP boundary, design and measured-cost budgets; verified before merge |
local-operator-interface-2026-09-13.md | Proposed 2026-09-13 for WI-100. Freshness-verified 0.5.0 assessment, local interface scope, shared state and execution boundaries; WI-101 separately evaluates operator workflow extensions and register admission |
tool-evaluation-2026-09-05.md | Assessed 2026-09-05 for WI-084. Implementation maturity, reproduced consumer gaps, prioritized improvements, repeated-pattern candidates and the case for a local inspector; recommendations, not implementation approval |
existing-promises-evidence-2026-09-06.md | Opened 2026-09-06 for WI-085. The reconciled baseline and one row per consumer promise the programme completes — authoritative claim, implementation path, the command actually run and its result, owning item, remaining limitation. Appended as items land |
existing-promises-remediation-prompt.md | Prepared 2026-09-05 at WI-084 closeout. Copy-ready implementation handoff for completing existing promises, with acceptance evidence and explicit separation from the assessment’s new-feature proposals |
product-brief.md | Written 2026-08-14. What it is · scaffold model · module boundary · output contract · CLI surface · upgrade story · non-goals · decisions |
module-catalog.md | Written 2026-08-14. The 15-module set: rung · deps · params · what each ships · install profiles · the corpus expectation matrix (Phase 6’s acceptance criterion) |
cli-surface.md | Never written. Planned for Phase 5 as detailed command behaviour beyond the brief’s summary; Phase 5 closed without it. The nine commands are specified in product-brief.md §6 and their current behaviour is rungs --help. Recorded rather than deleted — a planned document that was not needed is a finding about the plan |
parameters.md | Written 2026-08-15. What a module parameter is and how to set one. Deliberately lists none of them — rungs modules --params renders the list from the manifests, so there is no second inventory to go stale |
rift-forge-capability-matrix.md | Written 2026-08-15 for WI-031. Evidence-pinned portability decisions for the refreshed Rift Forge candidate; one adopted engine and explicit deferred adapter boundaries |
WI-032-claim-inventory.md | Written 2026-08-15 for WI-032. Source/date/command inventory for synchronized README, roadmap, module, and frontend claims |
cross-repo-evidence-2026-08-16.md | Written 2026-08-16 for WI-041. Whether rungs may hold evidence gathered across repositories — the case for at its strongest, the three things the phrase covers, and why only one is refused. Produced the amendment to ADR-0005 Tier C |
explain-census-2026-08-16.md | Written 2026-08-16 for WI-053. The measured false-positive rate of doctor --explain across every repository on this machine — 2,291 findings, 0 wrong — with the classifier’s own self-test, per-repo SHAs, and an explicit statement of what a one-operator corpus cannot prove |
external-review-2026-08-16b.md | Written 2026-08-16 for WI-048. The same reviewer’s re-assessment after WI-037 shipped, adjudicated against the working tree and the live site. Records the one claim that was wrong — the deployed site is current — and the two surfaces that genuinely were stale, which they missed |
external-review-2026-08-16.md | Written 2026-08-16 for WI-037. An outside review of the public repo and docs site, recorded in full and adjudicated claim by claim against the working tree — retained, partly, rejected, noted — including the two places it argues against an accepted ADR |
imperative-oracle-2026-09-06.md | Classified 2026-09-06 for WI-061, before any matcher existed. Every candidate line of the corpus read and classified rule / not-a-rule, per repository and commit. Its headline overturned the corpus document’s impression: 70% of the largest file’s candidates are project history. The detector is measured against this document |
imperative-corpus-2026-08-17.md | Gathered 2026-08-17 for WI-061. The corpus an imperative detector must be measured against, collected before any matcher exists — 102 local repos, per-repo counts because the file count is dominated by one project’s worktrees. Includes the two repos with zero candidates, and the finding that instruction-file scope has a far higher base rate than WI-042’s all-markdown scan. No detector has been written |
release-runbook.md | Written 2026-08-17 while preparing v0.2.0. The concrete release procedure for this repository — version decision, gating, changelog assembly, the surfaces no gate checks, and the irreversible tag/publish steps. Six traps, each one measured here rather than imagined. The portable procedure it defers to is cut-release |
release-readiness.md | Assessed 2026-08-15 for WI-033. Evidence record, not an approval: the dimension-by-dimension adoption assessment that found the clean-consumer bin failure v0.1.1 then fixed |
detection-verification.md | Detection results against all four source repos — Phase 6’s acceptance evidence |
web-design-system-prompt.md | Written 2026-08-14. Phase 7 working artifact, authoritative for nothing: the brief for a design system covering the landing page, the wiki, and contribute. The system it produces is what becomes authoritative |
Decisions live in ../decisions/. The one that settles the output
contract is ADR-0001.