savant-code 0.0.19 → 0.0.20

package/dev/test-prompts/archive/release-az-test-fid-085.md

@@ -1,306 +0,0 @@
-# Release A-Z Test — FID-085 Context Compaction System + message.content.map Regression Fix
-
-**Version:** v0.0.8
-**Purpose:** Regression and feature verification for the progressive context compaction system (FID-085) and the follow-up `message.content.map is not a function` runtime bug fix before release.
-
-**Ground Rules:**
-- Run from agent context (idle phase unless noted)
-- Do not require user interaction
-- Report pass/fail and any friction for every test
-- Write the final report to `dev/scratchpad/release-az-test-fid-085-report.md`
-
-**Available Tools:** read_files, glob, list_directory, spawn_agents, write_todos, basher, code_searcher
-
----
-
-## Tier 1: Build & Type Safety
-
-### T1.1 — Common workspace typecheck
-- Run `cd common && bun run typecheck`
-- **Expected:** exit code 0, no errors
-
-### T1.2 — Agent-runtime workspace typecheck
-- Run `cd packages/agent-runtime && bun run typecheck`
-- **Expected:** exit code 0, no errors
-
-### T1.3 — SDK workspace typecheck
-- Run `cd sdk && bun run typecheck`
-- **Expected:** exit code 0, no errors
-
-### T1.4 — CLI workspace typecheck
-- Run `cd cli && bun run typecheck`
-- **Expected:** exit code 0, no errors
-
-### T1.5 — Changed-file lint
-- Run ESLint with `--max-warnings 0` on the changed files:
-  - `packages/agent-runtime/src/context-compactor.ts`
-  - `packages/agent-runtime/src/run-agent-step.ts`
-  - `packages/agent-runtime/src/tools/tool-executor.ts`
-  - `packages/agent-runtime/src/tools/handlers/tool/run-readonly-command.ts`
-  - `common/src/types/session-state.ts`
-  - `common/src/constants/agents.ts`
-  - `common/src/util/messages.ts`
-  - `common/src/util/__tests__/messages.test.ts`
-  - `cli/src/utils/openrouter-models.ts`
-  - `cli/src/utils/create-run-config.ts`
-  - `cli/src/hooks/use-send-message.ts`
-  - `agents/savant/savant.ts`
-- **Expected:** zero warnings, zero errors
-
-### T1.6 — Messages unit tests
-- Run `bun test common/src/util/__tests__/messages.test.ts`
-- **Expected:** all tests pass
-
----
-
-## Tier 2: Context Compactor Components
-
-### T2.1 — ContextCompactor class exists
-- Read `packages/agent-runtime/src/context-compactor.ts`
-- Verify it exports a `ContextCompactor` class
-- Verify it exposes `microCompact`, `shouldAutoCompact`, `autoCompact`, and `reactiveCompact` methods
-- **Expected:** class is well-structured and matches the four-layer architecture
-
-### T2.2 — Compaction thresholds are reasonable
-- Read the `ContextCompactor` constructor and `getThresholds()`
-- Verify micro-compact keeps the 3 most recent tool results by default
-- Verify auto-compact threshold is `contextWindow - 30_000` tokens
-- **Expected:** defaults match the FID design
-
-### T2.3 — Compaction message type allows string content
-- Inspect the `CompactionMessage` interface
-- Verify `content` allows `string | Array<{ type: string; text?: string; [key: string]: unknown }>`
-- **Expected:** type is permissive enough to represent compacted placeholders
-
-### T2.4 — Placeholder shape after micro-compact
-- Read the `microCompact` implementation
-- Verify stale tool results are replaced with `content: [{ type: 'json', value: '[compacted]' }]`
-- **Expected:** no bare string content is emitted by the current code
-
-### T2.5 — ContextCompactor instantiation in the runtime
-- Read `packages/agent-runtime/src/run-agent-step.ts`
-- Find where `new ContextCompactor(...)` is created
-- Verify it is constructed with sensible thresholds (`maxContextLength`, `buffer`, `microCompactMaxKeepRecent`)
-- **Expected:** compactor is wired into the agent loop with correct defaults
-
----
-
-## Tier 3: Layer 2 — Micro-Compact
-
-### T3.1 — Micro-compact clears stale read results
-- Read `packages/agent-runtime/src/run-agent-step.ts`
-- Find where `contextCompactor.microCompact(...)` is called
-- Verify it runs before every API call in `loopAgentSteps`
-- **Expected:** stale read_files / code_search / glob results older than the 3 most recent are cleared
-
-### T3.2 — Micro-compact preserves recent tool results
-- Verify the logic keeps the `microCompactMaxKeepRecent` most recent tool results
-- **Expected:** recent results remain available to the model
-
-### T3.3 — Micro-compact reports tokens saved
-- Verify `microResult.tokensSaved` is logged or reported
-- **Expected:** user-facing message shows approximate tokens saved
-
----
-
-## Tier 4: Layer 3 — Auto-Compact
-
-### T4.1 — Auto-compact threshold check
-- Read `packages/agent-runtime/src/context-compactor.ts`
-- Verify `shouldAutoCompact` compares current token count to threshold
-- **Expected:** returns true when context exceeds `contextWindow - buffer`
-
-### T4.2 — Auto-compact circuit breaker
-- Verify the auto-compact circuit breaker: max 3 failures, 5-minute cooldown
-- **Expected:** repeated failures do not spam the API
-
-### T4.3 — Auto-compact integration in run-agent-step
-- Read `packages/agent-runtime/src/run-agent-step.ts`
-- Find the Layer 3 integration after each step completes
-- Verify it checks `contextCompactor.shouldAutoCompact(...)`
-- **Expected:** auto-compact is triggered at the right boundary
-
-### T4.4 — maxContextLength in AgentState
-- Read `common/src/types/session-state.ts`
-- Verify `AgentState` has a `maxContextLength` field (or equivalent)
-- **Expected:** field is present and used by the runtime
-
-### T4.5 — Savant reads maxContextLength
-- Read `agents/savant/savant.ts`
-- Verify `handleSteps` reads `agentState.maxContextLength`
-- **Expected:** agent is aware of its context budget
-
----
-
-## Tier 5: Layer 4 — Reactive Compact
-
-### T5.1 — Reactive compact triggers on prompt-too-long
-- Read `packages/agent-runtime/src/run-agent-step.ts`
-- Find the `reactiveCompact` call
-- Verify it triggers on an API `prompt-too-long` / content-length error
-- **Expected:** emergency truncation runs before giving up
-
-### T5.2 — Reactive compact preserves head and tail
-- Read the `reactiveCompact` implementation in `context-compactor.ts`
-- Verify it preserves the first message and the last 20% of messages
-- **Expected:** critical system prompt and recent context are retained
-
-### T5.3 — Reactive compact retries the API call
-- Verify the code retries the API call once after reactive compaction
-- **Expected:** one retry, then failure if it still does not fit
-
----
-
-## Tier 6: Message Conversion Regression (`message.content.map`)
-
-### T6.1 — Tool result with string content is coerced
-- Run `bun test common/src/util/__tests__/messages.test.ts`
-- Verify a tool message whose `content` is the bare string `[compacted]` is converted into a valid `tool-result` with `{ type: 'json', value: '[compacted]' }`
-- **Expected:** no `message.content.map is not a function` crash; output passes `modelMessageSchema`
-
-### T6.2 — System message with string content is coerced
-- Verify a system message whose `content` is a plain string is converted to a single string `content`
-- **Expected:** no `.map()` crash
-
-### T6.3 — User message with string content is coerced
-- Read `common/src/util/__tests__/messages.test.ts`
-- Find the user string-content test and run it via `bun test common/src/util/__tests__/messages.test.ts`
-- **Expected:** a user message whose `content` is a plain string is wrapped into a `TextPart[]` with no `.map()` crash
-
-### T6.4 — Assistant message with string content is coerced
-- Read `common/src/util/__tests__/messages.test.ts`
-- Find the assistant string-content test and run it via `bun test common/src/util/__tests__/messages.test.ts`
-- **Expected:** a raw assistant message whose `content` is a plain string is wrapped into a `TextPart[]` with no `.map()` crash
-
-### T6.5 — Invalid/null content falls back safely
-- Verify invalid `content` (e.g., `null` or `undefined`) falls back to an empty string for system messages and an empty array for user messages
-- **Expected:** no `.map()` crash
-
-### T6.6 — End-to-end compacted tool result
-- Run a small script that imports `convertCbToModelMessages` and `jsonToolResult` from the `@savant-code/common` package:
-  ```ts
-  import { convertCbToModelMessages, jsonToolResult } from '@savant-code/common/util/messages'
-  const messages = [{
-    role: 'tool' as const,
-    toolName: 'read_files',
-    toolCallId: 'call_compact',
-    content: jsonToolResult('[compacted]'),
-  }]
-  const result = convertCbToModelMessages({ messages, includeCacheControl: false })
-  process.exit(result.length > 0 && result[0].role === 'tool' ? 0 : 1)
-  ```
-- **Expected:** script exits 0, compacted tool results flow through conversion and reach the LLM safely
-
-### T6.7 — Defensive normalization does not mutate original messages
-- Pass a clone of a message array through `convertCbToModelMessages`
-- Verify the original array is unchanged
-- **Expected:** original messages are preserved
-
----
-
-## Tier 7: FSM & Tool Permission Bug Fixes
-
-### T7.1 — FSM phase ordering (BUG-004)
-- Read `packages/agent-runtime/src/tools/tool-executor.ts`
-- Search for `phase` ordering checks and `VALID_TRANSITIONS`
-- Verify `idle → red`, `red → green`, `green → audit`, and `audit → complete` transitions are allowed
-- **Expected:** no deadlock between idle → red → green → audit
-
-### T7.2 — run_readonly_command denylist (BUG-003)
-- Read `packages/agent-runtime/src/tools/handlers/tool/run-readonly-command.ts`
-- Search for `denylist` or `denyList` and `allowlist` or `allowList`
-- Verify the implementation uses a denylist (e.g., rejecting known write commands) rather than an allowlist
-- Search for `findstr` or `2>nul` in test files to confirm they work on Windows
-- **Expected:** more flexible read-only command execution
-
-### T7.3 — Agent ID in errors (BUG-001)
-- Search `packages/agent-runtime/src/tools/tool-executor.ts` for `agentId` or `agent.id` in thrown error strings
-- **Expected:** errors include the agent ID for easier debugging
-
-### T7.4 — devMode warning (BUG-006)
-- Search `packages/agent-runtime/src/tools/tool-executor.ts` for `devMode` and `warn`
-- Verify a warning is emitted when running in dev mode with tool errors
-- **Expected:** warning is present and helpful
-
-### T7.5 — ECHO_PROTOCOL_INSTRUCTIONS updated (BUG-005)
-- Read `common/src/constants/agents.ts`
-- Search for `ECHO_PROTOCOL_INSTRUCTIONS`
-- Verify the instructions mention the corrected FSM phase gating table
-- **Expected:** instructions reflect the latest protocol
-
----
-
-## Tier 8: Context Window Resolution
-
-### T8.1 — OpenRouter model context lengths (CTX-010)
-- Read `cli/src/utils/openrouter-models.ts`
-- Verify `inferContextLength` returns correct values for Grok (1M), GPT (256k), GLM (1M), MiniMax (256k)
-- **Expected:** no incorrect fallback to 128k
-
-### T8.2 — create-run-config passes contextWindow (CTX-007)
-- Read `cli/src/utils/create-run-config.ts`
-- Verify it passes `contextWindow` through
-- **Expected:** context window flows to run config
-
-### T8.3 — use-send-message resolves context window (CTX-007)
-- Read `cli/src/hooks/use-send-message.ts`
-- Verify `resolveContextWindowForModel` is wired through
-- **Expected:** selected model's context window is resolved
-
----
-
-## Tier 9: Documentation
-
-### T9.1 — FID status
-- Read `dev/fids/archive/FID-2026-0725-085-context-compaction-system.md`
-- Verify status is closed/archived and resolution fields are filled
-- **Expected:** ground-truth verification section complete
-
-### T9.2 — CHANGELOG entry
-- Read `CHANGELOG.md`
-- Verify a v0.0.8 entry exists for FID-085
-- **Expected:** entry matches the implementation
-
----
-
-## Tier 10: CLI Smoke (if tmux available)
-
-### T10.1 — CLI launches
-- If possible, launch the CLI with `bun run src/index.tsx --cwd ..` from `cli/`
-- Verify it starts without crashing
-- **Expected:** prompt appears
-
-### T10.2 — Compacted tool results do not crash conversion
-- Use the same snippet as T6.6 or run `bun test common/src/util/__tests__/messages.test.ts`
-- Verify no `message.content.map is not a function` error is thrown and the output array is non-empty
-- **Expected:** agent continues past compaction points without the `message.content.map` crash
-
----
-
-## Report Format
-
-After all tiers, write `dev/scratchpad/release-az-test-fid-085-report.md` with:
-
-1. **Executive Summary** — 3-5 sentences on release readiness
-2. **Tier-by-Tier Results** — For each test: Status, Notes, Friction Level (none/low/medium/high)
-3. **Blockers** — Any test that must be fixed before release
-4. **Pre-existing Issues** — Any failures not caused by this feature
-5. **Release Recommendation** — Go / No-Go with justification
-
----
-
-## Summary
-
-| Tier | Name | Tests | Purpose |
-|------|------|-------|---------|
-| 1 | Build & Type Safety | 6 | Does the code compile and pass tests? |
-| 2 | Context Compactor Components | 5 | Are the compactor classes present and typed correctly? |
-| 3 | Layer 2 — Micro-Compact | 3 | Does per-turn stale-result clearing work? |
-| 4 | Layer 3 — Auto-Compact | 5 | Does threshold-based summarization work? |
-| 5 | Layer 4 — Reactive Compact | 3 | Does emergency truncation work? |
-| 6 | Message Conversion Regression | 7 | Is the `message.content.map` crash fixed? |
-| 7 | FSM & Tool Permission Fixes | 5 | Are the 12 related bug fixes present? |
-| 8 | Context Window Resolution | 3 | Does the selected model's context window flow through? |
-| 9 | Documentation | 2 | Is the FID/CHANGELOG complete? |
-| 10 | CLI Smoke | 2 | Does the feature hold up in the real CLI? |
-| **Total** | | **41** | |

package/dev/test-prompts/archive/release-az-test-fid-087.md

@@ -1,219 +0,0 @@
-# Release A-Z Test — FID-087 Universal Copy Buttons
-
-**Version:** v0.0.9
-**Purpose:** Regression and feature verification for the universal copy button feature plus a quick sanity check of core CLI behavior before release.
-
-**Ground Rules:**
-- Run from agent context (idle phase unless noted)
-- Do not require user interaction
-- Report pass/fail and any friction for every test
-- Write the final report to `dev/scratchpad/release-az-test-fid-087-report.md`
-
-**Available Tools:** read_files, glob, list_directory, spawn_agents, write_todos, basher, code_searcher
-
----
-
-## Tier 1: Build & Type Safety
-
-### T1.1 — CLI typecheck
-- Run `cd cli && bun run typecheck`
-- **Expected:** exit code 0, no errors
-
-### T1.2 — Changed-file lint
-- Run ESLint on the changed files:
-  - `cli/src/components/blocks/copy-button.tsx`
-  - `cli/src/components/blocks/copyable-block.tsx`
-  - `cli/src/components/blocks/single-block.tsx`
-  - `cli/src/components/blocks/thinking-block.tsx`
-  - `cli/src/components/blocks/tool-branch.tsx`
-  - `cli/src/components/blocks/tool-block-group.tsx`
-  - `cli/src/components/blocks/agent-branch-wrapper.tsx`
-  - `cli/src/components/blocks/implementor-row.tsx`
-  - `cli/src/components/blocks/blocks-renderer.tsx`
-  - `cli/src/utils/clipboard.ts`
-- **Expected:** zero warnings, zero errors
-
-### T1.3 — Copy-button unit tests
-- Run `bun test src/__tests__/unit/copy-button.test.ts` from `cli/`
-- **Expected:** all tests pass
-
-### T1.4 — Clipboard utility tests
-- Run `bun test src/utils/__tests__/clipboard.test.ts` from `cli/`
-- **Expected:** all tests pass (note: one pre-existing environment-specific failure may occur when `TERM_PROGRAM=vscode` and SSH env vars are set)
-
-### T1.5 — Hook rules check
-- Read `cli/src/components/blocks/implementor-row.tsx`
-- Verify `useCallback`/`useState`/`useMemo` are called only at the top level of components, not inside `.map()` callbacks or loops
-- **Expected:** no hook calls inside loops or map callbacks
-
-### T1.6 — Import order / unused vars lint
-- Run `bun x eslint` on the changed files with `--max-warnings 0`
-- Verify no `import/order` warnings and no unused-variable warnings
-- **Expected:** zero warnings
-
----
-
-## Tier 2: Copy Button Components Exist
-
-### T2.1 — CopyButton component
-- Read `cli/src/components/blocks/copy-button.tsx`
-- Verify it exports a `CopyButton` component
-- Verify it uses the existing `Clickable` component
-- Verify it uses the restored `copyToClipboard` from `cli/src/utils/clipboard.ts`
-- **Expected:** no `navigator.clipboard`, no `theme.accent`, no absolute positioning
-
-### T2.2 — CopyableBlock component
-- Read `cli/src/components/blocks/copyable-block.tsx`
-- Verify it exports a `CopyableBlock` component
-- Verify it uses a flex column with a right-aligned row for the button
-- Verify it accepts an `isStreaming` prop that hides the button
-- **Expected:** OpenTUI-compatible layout, no absolute positioning
-
-### T2.3 — Clipboard utility
-- Read `cli/src/utils/clipboard.ts`
-- Verify original exports (`copyTextToClipboard`, `showClipboardMessage`, `subscribeClipboardMessages`, etc.) are still present
-- Verify the new `copyToClipboard` wrapper exists and suppresses global messages
-- **Expected:** old API preserved, new wrapper added
-
-### T2.4 — No invalid OpenTUI style props
-- Search `copy-button.tsx` and `copyable-block.tsx` for `position:`, `absolute`, `cursor:`, and `theme.accent`
-- **Expected:** none of these invalid/undefined style props are used
-
-### T2.5 — copyToClipboard failure contract
-- Temporarily force all copy methods to fail (e.g., register a renderer that throws and clear platform env)
-- Call `copyToClipboard('test')` and verify it returns `false` without throwing
-- **Expected:** returns `false`, no unhandled exception
-
----
-
-## Tier 3: Renderers Wired with CopyableBlock
-
-### T3.1 — SingleBlock
-- Read `cli/src/components/blocks/single-block.tsx`
-- Verify the text branch wraps non-user text in `CopyableBlock`
-- Verify it passes `isStreaming` to `CopyableBlock`
-- **Expected:** copy button appears on AI text blocks
-
-### T3.2 — ThinkingBlock
-- Read `cli/src/components/blocks/thinking-block.tsx`
-- Verify it wraps `<Thinking>` in `CopyableBlock`
-- Verify it hides the button while reasoning is incomplete
-- **Expected:** copy button appears on reasoning blocks after completion
-
-### T3.3 — ToolBranch
-- Read `cli/src/components/blocks/tool-branch.tsx`
-- Verify it wraps tool content in `CopyableBlock`
-- Verify it passes `isStreaming` to `CopyableBlock`
-- **Expected:** copy button appears on tool results blocks
-
-### T3.4 — ToolBlockGroup
-- Read `cli/src/components/blocks/tool-block-group.tsx`
-- Verify it wraps the group column in `CopyableBlock`
-- Verify it builds copy text from each `toolBlock`
-- **Expected:** copy button appears on grouped tool blocks
-
-### T3.5 — AgentBranchWrapper
-- Read `cli/src/components/blocks/agent-branch-wrapper.tsx`
-- Verify it wraps `<AgentBranchItem>` in `CopyableBlock`
-- Verify it passes `isStreaming` to `CopyableBlock`
-- **Expected:** copy button appears on agent branches
-
-### T3.6 — ImplementorGroup
-- Read `cli/src/components/blocks/implementor-row.tsx`
-- Verify each implementor column is wrapped in `CopyableBlock`
-- Verify the copy text includes implementor name, text, and tool outputs
-- **Expected:** copy button appears on implementor columns groups
-
----
-
-## Tier 4: Functional Checks
-
-### T4.1 — Copy text content
-- For each renderer, verify the `getCopyText` callback returns a non-empty string when content exists
-- Check that tool/agent/reasoning copy text includes a role prefix (`[Tool: ...]`, `[Agent: ...]`, `[Reasoning]`)
-- **Expected:** pasted content is readable and identifies the source
-
-### T4.2 — Streaming discipline
-- Verify `CopyableBlock` hides the button when `isStreaming` is true
-- Verify each parent passes the correct streaming flag
-- **Expected:** no copy button during streaming
-
-### T4.3 — Clipboard truncation
-- Verify the new `copyToClipboard` wrapper truncates text over 1MB with a `[truncated, see source]` suffix
-- **Expected:** large blocks do not crash the clipboard
-
----
-
-## Tier 5: Regression Checks
-
-### T5.1 — Existing copy button still works
-- Read `cli/src/components/copy-button.tsx`
-- Verify the existing inline copy button still exports `CopyButton`, `useCopyToClipboard`, and constants
-- **Expected:** no changes to the existing component
-
-### T5.2 — User content copy
-- Read `cli/src/components/blocks/user-content-copy.tsx`
-- Verify `UserBlockTextWithInlineCopy` is still used for user messages
-- **Expected:** user-message copy behavior unchanged
-
-### T5.3 — BlocksRenderer props
-- Read `cli/src/components/blocks/blocks-renderer.tsx`
-- Verify the previously unused `overrideCopyText` prop was removed
-- Verify no other props changed in a breaking way
-- **Expected:** cleaner interface
-
----
-
-## Tier 6: CLI Smoke (if tmux available)
-
-### T6.1 — CLI launches
-- If possible, launch the CLI with `bun run src/index.tsx --cwd ..` from `cli/`
-- Verify it starts without crashing
-- **Expected:** prompt appears
-
-### T6.2 — Copy buttons render
-- In the CLI, trigger a short agent response (e.g., "say hello")
-- Visually confirm a copy button appears next to the assistant text block
-- Click it and verify the clipboard contains the text
-- **Expected:** button visible and functional
-
----
-
-## Tier 7: Documentation
-
-### T7.1 — FID status
-- Read `dev/fids/FID-2026-0725-087-universal-copy-buttons.md`
-- Verify status is closed/archived and resolution fields are filled
-- **Expected:** ground-truth verification section complete
-
-### T7.2 — CHANGELOG entry
-- Read `CHANGELOG.md`
-- Verify a v0.0.9 entry exists for FID-087
-- **Expected:** entry matches the implementation
-
----
-
-## Report Format
-
-After all tiers, write `dev/scratchpad/release-az-test-fid-087-report.md` with:
-
-1. **Executive Summary** — 3-5 sentences on release readiness
-2. **Tier-by-Tier Results** — For each test: Status, Notes, Friction Level (none/low/medium/high)
-3. **Blockers** — Any test that must be fixed before release
-4. **Pre-existing Issues** — Any failures not caused by this feature
-5. **Release Recommendation** — Go / No-Go with justification
-
----
-
-## Summary
-
-| Tier | Name | Tests | Purpose |
-|------|------|-------|---------|
-| 1 | Build & Type Safety | 4 | Does the code compile and pass tests? |
-| 2 | Copy Button Components Exist | 3 | Are the new components present and correct? |
-| 3 | Renderers Wired | 6 | Is the affordance applied everywhere? |
-| 4 | Functional Checks | 3 | Does the copy behavior work correctly? |
-| 5 | Regression Checks | 3 | Did we break existing copy behavior? |
-| 6 | CLI Smoke | 2 | Does the feature render in the real CLI? |
-| 7 | Documentation | 2 | Is the FID/CHANGELOG complete? |
-| **Total** | | **23** | |