Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
/**
* Contract for the composer `@`/`/` mention popups
* (feat/composer-mentions, notes/composer-mentions-spec-2026-07-14.md).
* (feat/composer-mentions, docs/archive/composer-mentions-spec-2026-07-14.md).
*
* Pins the fragile ordering + SSR-safety guarantees:
* 1. The mention-popup keyboard branch runs BEFORE the Esc/drag branch and
Expand Down
11 changes: 11 additions & 0 deletions docs/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,8 +27,18 @@ This page is the authority map for Maka documentation. Code and contract tests r
- [Execution identity and evidence spine](./execution-evidence-spine.md)
- [AHE target protocol and evidence export](./ahe-target-protocol.md)
- [Skill catalog policy](./skill-catalog-policy.md)
- [Expert teams runtime](./expert-team-runtime.md)
- [Backend architecture chapters](./architecture/)

### Computer use

- [Foundation contract](./computer-use-foundation-contract.md)
- [Model-loop foundation](./computer-use-model-loop-foundation.md)
- [Evidence classes](./computer-use-evidence-classes.md)
- [Provider evidence contract](./computer-use-provider-evidence.md)
- [Host events contract](./computer-use-host-events-contract.md)
- [cua-driver artifact integrity](./cua-driver-artifact-integrity.md)

### Frontend and validation

- [Product design](../DESIGN.md)
Expand All @@ -54,3 +64,4 @@ This page is the authority map for Maka documentation. Code and contract tests r
- Keep progress and TODOs in issues or pull requests instead of copying them into stable documents.
- Move completed plans to `docs/archive/`; remove a document only after its unique references are updated or no longer needed.
- Prefer updating an existing authority over adding a parallel document.
- Keep PR follow-up records, incident investigations, and run logs in the pull request or `docs/archive/`, not as new current contracts. `notes/` and `docs/local/` are workspace-local scratch and stay untracked.
23 changes: 23 additions & 0 deletions docs/archive/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,9 @@ The root `DESIGN.md` owns shared product design intent. `docs/frontend-css-gover

Tracked `notes/` files were retired as an undocumented parallel authority. Point-in-time audits, migration plans, reference reverse-engineering, and design research remain here only for provenance. Current repository rules live in `ARCHITECTURE.md`, `DESIGN.md`, active cross-cutting contracts, local READMEs, source, and tests; workspace-specific agent instructions are supplied outside the tracked documentation set.

- `frontend-simplification-map-2026-07-13.md` — baseline knip/size measurement for the frontend simplification rounds landed in #871–#887.
- `composer-mentions-spec-2026-07-14.md` — the shipped v1 `@`/`/` composer mention-popup spec (#979); current behavior lives in `packages/ui` source and the composer-mention contract test.

## Security contract snapshots

- `memory-threat-model-pr-memory-1.md` — the contract-only PR-MEMORY-1 boundary, superseded as a description of the current product.
Expand All @@ -42,3 +45,23 @@ The dated runtime and sandbox files are completed execution plans. Each file rec
`runtime-v2-implementation-notes.md` records the initial Phase 1–4 runtime skeleton and is superseded by the backend architecture chapters and current source.

`agent-runtime-codex-sandbox-alignment.md` and `agent-runtime-codex-sandbox-todo.md` preserve the original sandbox discussion and phased checklist. Stable boundaries now live beside the runtime implementation; remaining work is tracked in issue #843.

## Computer use follow-up records

The computer-use follow-up files are per-PR follow-up records, incident investigations, and run logs from the #857–#985 delivery chain. Current computer-use authority lives in the contracts listed in `../README.md` and in source and contract tests.

- `computer-use-delivery-state.md` — `outcome_unknown` delivery-state follow-up; postcondition semantics are owned by the foundation contract and runtime tests.
- `computer-use-evidence-hardening.md` — fail-closed report hardening; the current boundary is owned by the provider evidence contract and evidence classes.
- `computer-use-executor-hardening.md` — post-merge review record of PR #893; durable decisions are folded into the foundation contract and the model-loop foundation, remaining work belongs in issues.
- `computer-use-runtime-hardening.md` — lifecycle fix record of PR #892; current fencing is owned by runtime source and tests.
- `computer-use-physical-input-guard.md` — physical-input interference incident and the guard policy it produced.
- `computer-use-process-restart-e2e.md` — process-restart soak goal, investigation, and run logs; the PID-reuse non-claim is restated in the host events contract.
- `computer-use-real-ax-model-e2e.md` — real-model AppKit AX harness record and provider run logs; current evidence classes and operator commands live in `../computer-use-evidence-classes.md`.

## Reverse-engineering records

- `expert-team-plan.md` and `expert-team-implementation.md` — WorkBuddy and QoderWork reverse-engineering and mapping notes behind the expert-teams feature. The shipped design contract is `../expert-team-runtime.md` (#971).

## Benchmark logs

- `economic-mechanisms-benchmark.md` — the first synthesis-cache benchmark entry for economic-mechanisms tracking issue #481 and slice #578. Current progress belongs in those issues.
File renamed without changes.
2 changes: 2 additions & 0 deletions docs/computer-use-foundation-contract.md
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,8 @@

`PASS`:当前证据直接覆盖;`PARTIAL`:组件证据存在但 production 闭环不足;`FAIL`:当前实现违反合同;`UNKNOWN`:缺少足够证据。

本矩阵记录 #857 拆分链建立时的基线状态,用于界定各 stacked PR 的验证责任。拆分链合入后,各领域当前状态以源码与合同测试为准。

| Contract area | 状态 | 当前证据 | 拆分链需要的证据 |
|---|---|---|---|
| Frame/window binding、duplicate rejection | PASS | frame state、bound-action、stale/duplicate tests | 在 Runtime slice 保留 focused tests |
Expand Down
2 changes: 1 addition & 1 deletion docs/expert-team-runtime.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
Expert teams let a **lead** persona fan a task out to specialist **member** experts,
each running as a tool-scoped child agent, then synthesize their results. It is the
star topology reverse-engineered from WorkBuddy's `team` experts and QoderWork's
sub-agent pipelines (see [expert-team-implementation.md](expert-team-implementation.md)),
sub-agent pipelines (see [expert-team-implementation.md](archive/expert-team-implementation.md)),
rebuilt entirely on Maka's existing child-agent machinery — no new orchestration
engine, no mesh mailbox, no shared task board.

Expand Down
2 changes: 1 addition & 1 deletion packages/runtime/src/__tests__/ai-sdk-backend.test.ts
Original file line number Diff line number Diff line change
Expand Up @@ -1089,7 +1089,7 @@ describe('AiSdkBackend model history', () => {
// is the synthesis cache's replay-token saving — no live model or network,
// so the number is reproducible in CI. (Live Terminal-Bench runs can't drive
// this: their single-long-turn shape never archives-then-retrieves a turn, so
// the write path stays source_missing — see docs/economic-mechanisms-benchmark.md.)
// the write path stays source_missing — see docs/archive/economic-mechanisms-benchmark.md.)
const CHARS_PER_TOKEN = 1; // 1:1 so prompt chars are directly the token estimate
const archivedResult = { body: 'RAW_SYNTHESIS_ARCHIVE_PAYLOAD '.repeat(80).trim() };
const serialized = JSON.stringify(archivedResult);
Expand Down
2 changes: 1 addition & 1 deletion packages/ui/src/chat-input-behavior.ts
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,7 @@ export function focusTextInputAtEnd(input: TextInputSelectionTarget): void {
* Composer `@` / `/` mention trigger detection.
*
* Decompiled from the QoderWork/WorkBuddy composer bundles and adapted to our
* plain-text v1 (see notes/composer-mentions-spec-2026-07-14.md). Pure so it
* plain-text v1 (see docs/archive/composer-mentions-spec-2026-07-14.md). Pure so it
* can be unit-pinned without a DOM: given the current textarea value + caret
* offset it reports the active trigger, the query typed after it, and the
* trigger char's index (so the caller can splice `[start, caret)` on select).
Expand Down
2 changes: 1 addition & 1 deletion packages/ui/src/composer-mention-popup.tsx
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
/**
* Composer `@` file / `/` skill mention popup — a presentational overlay
* rendered by Composer while a trigger is active. See
* notes/composer-mentions-spec-2026-07-14.md for the v1 plain-text model.
* docs/archive/composer-mentions-spec-2026-07-14.md for the v1 plain-text model.
*
* All state (which trigger, the filtered items, the highlighted index) lives in
* Composer; this component only paints the list and forwards hover/click to
Expand Down
2 changes: 1 addition & 1 deletion packages/ui/src/composer.tsx
Original file line number Diff line number Diff line change
Expand Up @@ -232,7 +232,7 @@ export const Composer = forwardRef<
onPermissionModeChange?(mode: PermissionMode): void | Promise<void>;
/**
* Composer mention popups (v1 plain-text tokens; see
* notes/composer-mentions-spec-2026-07-14.md). Both are optional and the
* docs/archive/composer-mentions-spec-2026-07-14.md). Both are optional and the
* whole feature no-ops when absent (SSR contracts render Composer with
* minimal props):
* - `mentionSkills` powers the `/` popup — pass only ENABLED skills; the
Expand Down
Loading