savant-code 0.0.17 → 0.0.19

package/dev/test-prompts/archive/comprehensive-az-test-v6.md

@@ -0,0 +1,631 @@
+# Savant-Code — Comprehensive A-Z System Test v6 (Official)
+
+**Purpose:** Exhaustive functional test of every tool, agent, FSM gate, slash command, skill, SDK behavior, provider integration, path safety, TUI component, and CLI interaction in the Savant-Code harness. This is the official regression suite; run it after every significant change or before a release.
+
+**Mode:** Interactive live execution inside the Savant CLI. You MUST call every tool, agent, and slash command listed below. Report PASS/FAIL for each with evidence. Do not skip any item. If a tool or agent fails, capture the exact error message and continue testing the rest.
+
+**Environment:** The test runs inside the Savant CLI with ECHO Protocol v0.2.0 active.
+
+**Platform notes:**
+- Production runs on Linux — all tests should pass cleanly on Linux.
+- Windows local dev has known pre-existing platform test infrastructure issues (see FID-015) — some SDK tool tests fail on Windows due to mock fs key mismatch. Production behavior is correct.
+- For CI, run on Linux runners.
+- Run on a clean working tree if possible; document any uncommitted changes in the report.
+
+**Before you start:**
+1. Confirm the CLI boots without errors.
+2. Confirm you can see the right sidebar with the current FSM phase.
+3. Keep a scratchpad note of every command/output for the final report.
+
+---
+
+## Phase 1: Boot & Identity
+
+### Test 1: ECHO Protocol bootstrap
+```bash
+/scratchpad echo-bootstrap-check
+```
+**Expected:** Output confirms ECHO.md loaded, version v0.2.0 active, strict_mode: true, language: typescript.
+
+### Test 2: Open FIDs scan
+```bash
+/scratchpad fid-scan
+```
+**Expected:** Lists all open FIDs in `dev/fids/` (excluding archived). Verify at least one open FID or confirm the list is empty (and why).
+
+### Test 3: Phase display
+```bash
+/phase
+```
+**Expected:** Returns current FSM phase. Should be `idle` when no work is in progress.
+
+### Test 4: Model metadata awareness (FID-054)
+Start a new chat and ask the agent: "What model are you running on, and what is its context window?"
+**Expected:** The agent reports the actual model selected in the CLI/model picker, including context-window, provider, and release info. It should NOT claim a hardcoded model such as `anthropic/claude-opus-4.8`.
+
+---
+
+## Phase 2: Direct Tools
+
+### Test 5-14: Read tools (10 items)
+```bash
+/read_files ECHO.md
+/read_files protocol.config.yaml
+/read_files package.json
+/read_files ARCHITECTURE.md
+/read_subtree cli/src/components
+/list_directory dev/fids
+/list_directory agents
+/glob "agents/**/*.ts"
+/code_search "resolveAndContain" common/src/util
+/code_search "fsmPhase" packages/agent-runtime/src
+```
+
+### Test 15-19: Write tools (FID-bound)
+**Prerequisite:** Transition to GREEN phase. Open FID first.
+```bash
+/transition red
+/transition green  # Requires open FID
+/write_file test.txt "hello"  # Should succeed in GREEN
+/read_files test.txt
+/transition audit
+/rm test.txt  # Bash allowed in AUDIT
+```
+
+### Test 20-24: FSM transitions
+```bash
+/transition idle→red        # Should succeed
+/transition red→green       # Should succeed (open FID exists)
+/transition green→audit     # Should succeed
+/transition audit→complete  # Should succeed
+/transition idle→audit      # Should FAIL: "INVALID FSM transition"
+```
+
+### Test 25-27: Illegal FSM transitions
+```bash
+/transition idle→green      # Should FAIL
+/transition audit→green     # Should FAIL
+/transition complete→red    # Should FAIL
+```
+
+---
+
+## Phase 3: Dev Override (FID-003 / v4 update)
+
+> **Note:** `/dev` no longer requires a passphrase as of this test version.
+
+### Test 28-31: Dev mode activation (no password)
+```bash
+/dev on                    # Activates dev override immediately
+/transition idle→green     # Should succeed without open FID (dev override bypass)
+/write_file test-dev.txt "x"  # Should succeed in any phase
+/dev off                   # Returns to normal FID-bound mode
+/transition green          # Should FAIL (no open FID + dev mode off)
+```
+
+### Test 32: Unknown /dev subcommand
+```bash
+/dev password              # Should report unknown subcommand and suggest "/dev on" / "/dev off"
+```
+
+### Test 33: /dev idempotency
+```bash
+/dev on    # Should activate dev override
+/dev on    # Should report "Dev override is already active."
+/dev off   # Should deactivate dev override
+/dev off   # Should report "Dev override is already off."
+```
+
+---
+
+## Phase 4: Slash Commands
+
+### Test 33-36: ECHO commands
+```bash
+/fids       # Lists open FIDs
+/fid <any-open-FID>  # Shows FID details
+/phase      # Shows current FSM phase
+/phase red  # Transitions to red
+```
+
+### Test 37: Model command
+```bash
+/model  # Opens the model picker if the live catalog loaded; otherwise shows current model or a fallback message
+```
+**Expected:** Either a picker appears, or a system message shows the current model and a helpful fallback.
+
+### Test 38: Bare /dev toggle
+```bash
+/dev      # With no argument, should activate dev override (same as "/dev on")
+/dev off  # Deactivate
+```
+**Expected:** Bare `/dev` behaves as `/dev on` when dev mode is off.
+
+---
+
+## Phase 5: Agent Roster (FID-006)
+
+### Test 39-47: 9 agents
+For each agent below, verify it exists and has the correct tool set per `ARCHITECTURE.md`:
+
+| # | Agent | Verify |
+|---|-------|--------|
+| 1 | Orchestrator | Has spawn_agents, read_files, transition_phase; in Hybrid Mode also has write_file, str_replace; but NOT bash |
+| 2 | Detective | Has code_search, set_output; NOT write tools |
+| 3 | Forge | Has write_file, str_replace; NOT bash |
+| 4 | Verifier | No write tools (reads only); has Audit Checklist in prompt |
+| 5 | Recorder | Has write_file, read_files, glob, grep, set_output; has transition_phase |
+| 6 | Thinker | Has sequentialthinking |
+| 7 | Scout | Has spawn_agents; no write tools |
+| 8 | Researcher | Has web_search, read_url, read_docs; no write tools |
+| 9 | Scribe | Has read_files, write_file, glob, grep, set_output |
+
+```bash
+/code_search "Orchestrator" agents/base2/base2.ts
+/code_search "Detective" agents/detective/detective.ts
+# ... etc for all 9
+```
+
+---
+
+## Phase 6: Scout file-finding (FID-007 F-A)
+
+### Test 47-49: Glob behavior
+```bash
+/scout "find auth files"
+/scout "locate test prompts"
+/scout "search FID docs"
+```
+
+---
+
+## Phase 7: MCP proxy timeout (FID-007 F-B)
+
+### Test 50-52: Timeout behavior
+```bash
+/code_search "withTimeout" common/src/mcp/client.ts
+/code_search "MCP_TIMEOUT" common/src/util/protocol-config.ts
+# Verify 2-second timeout for first byte, 30-second for completion
+```
+
+---
+
+## Phase 8: FSM phase inheritance (FID-004)
+
+### Test 53-55: Subagent inheritance
+```bash
+/code_search "fsmPhase" packages/agent-runtime/src/tools/handlers/tool/spawn-agent-utils.ts
+/code_search "createAgentState" packages/agent-runtime/src/tools/handlers/tool/spawn-agent-utils.ts
+# Verify subagents inherit fsmPhase from parent
+```
+
+---
+
+## Phase 9: Perfection Loop + circuit breaker
+
+### Test 56-59: FSM gates
+```bash
+# Test 10-iteration circuit breaker
+# Open 10+ FIDs and cycle them through to trigger hard stop
+```
+
+---
+
+## Phase 10: FID-013 v3 path safety
+
+### Test 60-64: Path safety
+```bash
+/code_search "resolveAndContain" common/src/util/paths.ts
+/code_search "resolveAndContain" packages/agent-runtime/src/tools/handlers/tool/write-file.ts
+/code_search "resolveAndContain" packages/agent-runtime/src/tools/handlers/tool/str-replace.ts
+/code_search "resolveAndContain" packages/agent-runtime/src/tools/handlers/tool/apply-patch.ts
+/code_search "resolveAndContain" packages/agent-runtime/src/tools/tool-executor.ts
+```
+
+---
+
+## Phase 11: FID-014 v2 SDK-side realpath
+
+### Test 65-69: SDK-side realpath wiring
+```bash
+/code_search "resolveAndContain" sdk/src/tools/change-file.ts
+/code_search "resolveAndContain" sdk/src/tools/apply-patch.ts
+/code_search "realpathFn" common/src/util/paths.ts
+/code_search "realpathFn" sdk/src/tools/change-file.ts
+/code_search "realpathFn" sdk/src/tools/apply-patch.ts
+# Verify realpathFn injection in SDK tools
+```
+
+### Test 70-72: SDK path safety test coverage
+```bash
+# Verify test #7 in change-file.test.ts asserts path-escape rejection
+/scratchpath read-test "sdk/src/__tests__/change-file.test.ts"
+# Look for "rejects absolute paths outside the project" test
+```
+
+---
+
+## Phase 12: Skills system (FID-002)
+
+### Test 73-80: 7 coding standards as skills
+```bash
+/list_directory .agents/skills
+/code_search "skill" cli/src/chat.tsx | head -10
+/code_search "skill" cli/src/utils/settings.ts | head -10
+# Verify all 7 skills present: coding-typescript, coding-python, coding-rust, coding-java, coding-go, coding-csharp, release-workflow
+```
+
+---
+
+## Phase 13: CLI/TUI edge cases (Master TUI Rebuild)
+
+### Test 81-87: TUI behavior
+- Open chat → see right sidebar with FSM phase
+- Type a message → submit
+- Press Ctrl+C → graceful exit
+- Press `/` → slash command menu appears (CommandPalette)
+- Navigate history with arrow keys
+- Test tab completion for paths
+- Verify tokens update in right sidebar (ContextWindow: 200k)
+
+### Test 88: Command palette (FID-033d)
+In the chat input type `/` or a slash command prefix.
+**Expected:** The native OpenTUI `<select>` command palette renders inline above the input without hiding the input. Escape closes it.
+
+### Test 89: Toast system (FID-033d)
+Trigger an action that should produce a toast (e.g., invalid /dev subcommand).
+**Expected:** A toast appears in the bottom-right, can be dismissed, and does not stack beyond `MAX_TOASTS`.
+
+### Test 90: Theme toggle
+```bash
+/theme:toggle
+```
+**Expected:** Theme switches between dark and light; sidebar and syntax highlighting update accordingly.
+
+---
+
+## Phase 14: Knowledge files (FID-005)
+
+### Test 91-95: LEARNINGS wiring
+```bash
+/read_files dev/LEARNINGS.md
+/code_search "LEARNINGS" common/src/util/strings.ts
+/code_search "KNOWLEDGE_FILE_NAMES" common/src/util/strings.ts
+# Verify LEARNINGS.md in knowledge pipeline
+```
+
+---
+
+## Phase 15: Typecheck + Tests
+
+### Test 96-100: Build state
+```bash
+cd sdk && bun run typecheck       # Should pass
+cd common && bun run typecheck    # Should pass
+cd packages/agent-runtime && bun run typecheck  # Should pass
+cd cli && bun run typecheck       # Should pass
+bun test common/src/util/__tests__/paths.test.ts  # 18 pass / 4 skip / 0 fail
+```
+
+---
+
+## Phase 16: Rebrand readiness check
+
+### Test 101-105: Branding consistency
+```bash
+# Verify "Savant" branding throughout
+/code_search "Savant" cli/src/ --type tsx | wc -l
+/code_search "savant-free" cli/src/ --type tsx | wc -l
+# Savant count should be >> savant-free count (rebrand complete)
+
+# Verify ECHO Protocol references
+/code_search "ECHO" ECHO.md | head -5
+/read_files dev/CHANGELOG.md  # Check for "Savant" branding in recent entries
+```
+
+---
+
+## Phase 17: Code-Reviewer Agent Spawn Frequency (FID-057)
+
+**Purpose:** Determine whether the Verifier agent is actually spawned during normal code-change flows based on the new objective trigger criteria.
+
+### Test 106: Search for reviewer spawns in source
+```bash
+/code_search "spawn_agents" packages/agent-runtime/src/tools/handlers/tool/spawn-agents.ts
+/code_search "code-reviewer" agents/ cli/src/ packages/agent-runtime/src/ -g *.ts -n
+/code_search "verifier" agents/savant/savant.ts
+```
+**Expected evidence to capture:**
+- List every location where any `verifier` or `code-reviewer-*` pattern is referenced.
+- Note which reviewer identifiers actually exist in the codebase.
+- Note the objective trigger criteria: 10+ lines, 2+ files, new API, security, user request, Forge usage.
+
+### Test 107: Trigger a code change and observe spawned subagents
+Perform a trivial, safe edit via the agent (e.g., write a comment to a scratch file in `dev/scratchpad/`).
+**Expected:** The agent's own step summary lists every spawned subagent. Record whether the Verifier agent appeared.
+
+### Test 108: Document policy recommendation
+If the Verifier agent is not being spawned for trivial changes, record: "Verifier agent correctly skipped for trivial changes (< 10 lines, single file, no new imports)."
+If it is spawned, record the exact trigger condition and frequency.
+
+---
+
+## Phase 18: Provider integration (FID-054)
+
+### Test 109-111: OpenRouter gateway cache
+```bash
+# Immediately after CLI boot, the gateway catalog should be warming in the background
+/code_search "fetchGatewayModels" cli/src/index.tsx
+/code_search "getCachedGatewayModels" cli/src/utils/openrouter-models.ts
+```
+**Expected:** Boot calls `fetchGatewayModels()` non-blockingly; `use-send-message.ts` resolves the effective model and looks it up in the cached catalog.
+
+---
+
+## Phase 19: Hybrid Mode (FID-002, FID-003)
+
+**Purpose:** Verify that Savant can write code directly without spawning Forge, and that Forge is only used for complex tasks.
+
+### Test 112: Hybrid Mode system prompt
+```bash
+/code_search "primary coder" agents/savant/savant.ts
+/code_search "write code directly" agents/savant/savant.ts
+```
+**Expected:** System prompt says "You are the primary coder — write code directly using write_file and str_replace."
+
+### Test 113: Hybrid Mode instructions
+```bash
+/code_search "Hybrid Mode" agents/savant/savant.ts
+/code_search "write ALL code changes directly" agents/savant/savant.ts
+```
+**Expected:** Instructions say "Write ALL code changes directly using write_file and str_replace" for most tasks.
+
+### Test 114: Forge only for complex tasks
+```bash
+/code_search "Spawn Forge only" agents/savant/savant.ts
+/code_search "> 3 files AND requires new imports" agents/savant/savant.ts
+```
+**Expected:** Forge is only spawned for complex changes (> 3 files + new APIs, novel architecture, verification fails twice, user requests Forge).
+
+### Test 115: Direct writing test
+Ask the agent: "Write a simple comment to test-sandbox/src/comment.ts saying '// Hybrid mode test'"
+**Expected:** Agent writes the file directly using `write_file` without spawning Forge.
+
+---
+
+## Phase 20: Verifier Trigger Criteria (FID-057)
+
+**Purpose:** Verify the Verifier agent is triggered based on objective, measurable criteria instead of subjective judgment.
+
+### Test 116: Objective trigger criteria in prompt
+```bash
+/code_search "Verifier trigger" agents/savant/savant.ts
+/code_search "objective criteria" agents/savant/savant.ts
+```
+**Expected:** Prompt contains: "Spawn the Verifier to review code changes when ANY of these apply: (1) change is 10+ lines, (2) change touches 2+ files, (3) new function or API added, (4) security-sensitive code touched, (5) user explicitly requests review, (6) when Forge was used to implement changes."
+
+### Test 117: Skip criteria
+```bash
+/code_search "Skip Verifier only" agents/savant/savant.ts
+```
+**Expected:** "Skip Verifier only when change is < 10 lines AND single file AND no new imports."
+
+### Test 118: noReview flag gating
+```bash
+/code_search "!noReview &&" agents/savant/savant.ts
+```
+**Expected:** The Verifier trigger instruction is gated behind `!noReview` so fast mode skips it.
+
+### Test 119: Trivial change test
+Ask the agent: "Fix the typo in test-sandbox/src/comment.ts — change 'Hybird' to 'Hybrid'"
+**Expected:** Agent fixes the typo (< 10 lines, single file, no new imports) WITHOUT spawning Verifier.
+
+### Test 120: Non-trivial change test
+Ask the agent: "Add a new function `calculateSum` to test-sandbox/src/utils.ts that takes an array of numbers and returns their sum, with proper error handling and tests"
+**Expected:** Agent spawns Verifier after implementation (new function added).
+
+---
+
+## Phase 21: Audit Checklist in Verifier (FID-057)
+
+**Purpose:** Verify the Verifier agent checks against the ECHO Audit Checklist.
+
+### Test 121: Audit Checklist in Verifier prompt
+```bash
+/code_search "ECHO Audit Checklist" agents/verifier/verifier.ts
+/code_search "No magic numbers" agents/verifier/verifier.ts
+/code_search "Law 14" agents/verifier/verifier.ts
+/code_search "Law 6" agents/verifier/verifier.ts
+/code_search "Law 5" agents/verifier/verifier.ts
+```
+**Expected:** Verifier's instructionsPrompt contains the 6-item ECHO Audit Checklist:
+- No magic numbers or strings (all constants extracted)
+- All names follow language conventions
+- Error handling is comprehensive (Law 14)
+- No type safety shortcuts (Law 6)
+- No TODOs without FID references (Law 5)
+- Implementation matches converged FID spec (if applicable)
+
+### Test 122: Checklist items not duplicated with Guidelines
+```bash
+/code_search "dead code" agents/verifier/verifier.ts
+/code_search "missing imports" agents/verifier/verifier.ts
+```
+**Expected:** "dead code" and "missing imports" appear only in the Guidelines section, NOT in the Audit Checklist (avoiding duplication).
+
+---
+
+## Phase 22: Batch Operations (FID-058)
+
+**Purpose:** Verify the agent batches multiple file edits before running verification.
+
+### Test 123: Batch operations instruction
+```bash
+/code_search "Batch operations" agents/savant/savant.ts
+/code_search "write ALL files first" agents/savant/savant.ts
+/code_search "run typecheck/lint ONCE" agents/savant/savant.ts
+```
+**Expected:** Instructions say: "When making multiple related file changes, write ALL files first, then run typecheck/lint ONCE at the end."
+
+### Test 124: Batch operations test
+Ask the agent: "Create test-sandbox/src/math.ts with add, subtract, multiply functions and test-sandbox/src/math.test.ts with tests for all three"
+**Expected:** Agent writes both files, then runs verification once (not after each file).
+
+---
+
+## Phase 23: Smart Phase Transitions (FID-059)
+
+**Purpose:** Verify the agent can skip phases when appropriate.
+
+### Test 125: Smart Phase Transitions section
+```bash
+/code_search "Smart Phase Transitions" agents/savant/savant.ts
+/code_search "Skip When" agents/savant/savant.ts
+```
+**Expected:** Section contains skip-when table for RED, GREEN deliberation, and Full AUDIT phases.
+
+### Test 126: Law 3 never skipped
+```bash
+/code_search "Law 3 is NEVER skipped" agents/savant/savant.ts
+```
+**Expected:** "Law 3 (Verify Before Proceed) is NEVER skipped — verification always happens."
+
+### Test 127: Skip RED when issues known
+```bash
+/code_search "Issues already known" agents/savant/savant.ts
+```
+**Expected:** RED can be skipped when "Issues already known from prior analysis."
+
+### Test 128: Skip GREEN deliberation for obvious fixes
+```bash
+/code_search "Fix is obvious" agents/savant/savant.ts
+```
+**Expected:** GREEN deliberation can be skipped when "Fix is obvious (typo, missing import, constant change)."
+
+### Test 129: Skip AUDIT for trivial changes
+```bash
+/code_search "Change is < 10 lines" agents/savant/savant.ts
+```
+**Expected:** Full AUDIT can be skipped when "Change is < 10 lines AND single file AND typecheck/lint already pass inline."
+
+---
+
+## Phase 24: Parallel Agent Batching (FID-060)
+
+**Purpose:** Verify the agent fires all independent agents in a single spawn_agents call.
+
+### Test 130: Parallel agent batching instruction
+```bash
+/code_search "Parallel agent batching" agents/savant/savant.ts
+/code_search "fire them ALL in a single" agents/savant/savant.ts
+/code_search "Promise.allSettled" agents/savant/savant.ts
+```
+**Expected:** Instructions say: "When spawning multiple agents that don't depend on each other, fire them ALL in a single spawn_agents call — they run in parallel via Promise.allSettled."
+
+### Test 131: Dependency table
+```bash
+/code_search "Independent agents" agents/savant/savant.ts
+/code_search "Dependent agents" agents/savant/savant.ts
+```
+**Expected:** Table shows: Independent (Detective + Researcher + Thinker), Dependent (Scout waits for Detective, Forge waits for Thinker, Verifier waits for Forge).
+
+### Test 132: Sequencing guidance updated
+```bash
+/code_search "Sequence agents when needed" agents/savant/savant.ts
+```
+**Expected:** "Only sequence agents when there are data dependencies" (not "Sequence agents properly").
+
+---
+
+## Phase 25: Double Audit Enforcement (FID-057)
+
+**Purpose:** Verify Hybrid Mode satisfies the Double Audit requirement via two independent methods.
+
+### Test 133: Double Audit documentation in ECHO.md
+```bash
+/code_search "Double Audit" ECHO.md
+/code_search "bashers.*static analysis" ECHO.md
+/code_search "Verifier.*code review" ECHO.md
+```
+**Expected:** ECHO.md documents: "Method 1: bashers (typecheck/lint) — static analysis" and "Method 2: Verifier — independent code review."
+
+### Test 134: Self-reporting prohibition
+```bash
+/code_search "Self-reporting is prohibited" ECHO.md
+```
+**Expected:** "Self-reporting is prohibited. The Orchestrator that writes code must not be the one to verify it."
+
+### Test 135: Verifier trigger criteria in ECHO.md
+```bash
+/code_search "Verifier Trigger Criteria" ECHO.md
+/code_search "objective" ECHO.md
+```
+**Expected:** ECHO.md contains the objective Verifier trigger criteria table matching savant.ts.
+
+---
+
+## Phase 26: bun dev startup (FID-055)
+
+**Purpose:** Verify the CLI starts without errors after the unescaped backticks fix.
+
+### Test 136: bun dev starts successfully
+```bash
+bun dev
+```
+**Expected:** CLI boots without "Expected ';'" errors. Prebuild step completes successfully.
+
+### Test 137: Prebuild agents step
+```bash
+/code_search "prebuild-agents" scripts/
+```
+**Expected:** The prebuild step (`scripts/prebuild-agents.ts`) runs without template literal syntax errors.
+
+---
+
+## Reporting
+
+After completing all tests, write a comprehensive report to `dev/scratchpad/2026-<current-month>-<current-day>-final-rebrand-qa-report.md` (substitute the current month and day) with:
+- Test ID, Status (PASS/FAIL), Evidence for each item
+- Summary section with total pass/fail counts
+- Recommendations for any failures
+- Sign-off if all critical tests pass
+
+**Critical success criteria:**
+- All typechecks pass (zero errors)
+- All 9 agents present with correct tool sets
+- All FSM transitions work as expected
+- `/dev` activates and deactivates without a password
+- Code-reviewer agent spawn behavior is documented
+- All 3 FID-014 v2 fixes verified in source
+- SDK-side realpath defense wired in 2 SDK files
+- Cross-platform path normalization in paths.ts
+- OpenRouter model metadata is dynamically injected into the orchestrator prompt (no hardcoded model claim)
+- Hybrid Mode works (Savant writes code directly)
+- Verifier trigger uses objective criteria
+- Audit Checklist present in Verifier prompt
+- Batch operations instruction present
+- Smart Phase Transitions section present
+- Parallel Agent Batching instruction present
+- Double Audit enforced in Hybrid Mode
+- bun dev starts without errors
+
+**Acceptable caveats (document in report, don't fail):**
+- Windows local dev has 18 SDK tool test failures (pre-existing platform test infra issues) — tracked as FID-015
+- Token tracking in UI may not update in real-time (known UX issue)
+
+---
+
+## Agent input
+
+When you complete your entire run, create another report from **YOUR viewpoint** as the operator. Save it to `dev/scratchpad/agent-feedback-<YYYYMMDD>.md` and include:
+
+1. **Pain points:** Anything that slowed you down, confused you, or felt broken.
+2. **Tool failures:** Which tools failed, why, and what error text you saw. Include exact messages.
+3. **Wasted steps:** Commands or tool calls that gave no useful information.
+4. **Missing context:** Files, docs, or state you wish had been available up front.
+5. **Workflow friction:** Anything about the FSM, slash commands, or UI that made testing harder.
+6. **Quality of outputs:** Which outputs were excellent, which were noisy, and which were wrong.
+7. **Suggestions:** Concrete changes to prompts, tools, or the CLI that would make your job easier.
+8. **One thing to add/remove to this test prompt:** Based on this run, what single test or check should be added, removed, or changed?
+
+Be honest and detailed. We will iterate until the workflow is perfect for both operator and user.