savant-code 0.0.19 → 0.0.20

package/dev/test-prompts/archive/goal-loop-cli-test.md

@@ -1,142 +0,0 @@
-# CLI Test Prompt: Goal/Loop System End-to-End
-
-## Purpose
-Validate that the `/goal` and `/loop` slash commands work end-to-end,
-including scheduling, goal evaluation, sidebar status, and CLI output.
-
-## Live Run Prompt (paste into the CLI)
-
-```
-Run a full goal/loop smoke test in this project:
-1. Set a goal: /goal typecheck and lint pass with no new errors
-2. Start a 30-second loop: /loop 30s run typecheck and report the result
-3. Confirm the right sidebar shows the active loop with cadence "30s"
-4. Run /loop status and verify it shows the schedule and next run time
-5. Wait for the next loop tick (~30 seconds), verify the agent sends another message, then run /loop stop
-6. Confirm the sidebar returns to "No active loop"
-```
-
-## Prerequisites
-- The SavantCode CLI is running in an interactive tmux session
-- The agent runtime is connected and ready
-
----
-
-## Test 1: /goal command
-
-### Step 1 — Set a goal
-Type the following in the CLI:
-```
-/goal all tests pass with no new failures
-```
-
-### Expected Result
-- The CLI shows a confirmation message with the goal condition
-- The goal condition is injected into the system message for the agent runtime
-- A new message appears in the chat listing the active goal
-
-### Verified
-- [ ] Goal confirmation message appears
-- [ ] Goal condition is visible in system message injection
-- [ ] Goal is listed in the chat as an active goal
-
----
-
-## Test 2: /loop command — start
-
-### Step 2 — Start a loop
-Type the following in the CLI:
-```
-/loop 1m review and fix any failing tests
-```
-
-### Expected Result
-- The CLI shows a "Loop started" confirmation with cadence and prompt
-- The first run executes immediately (prompt is sent to the agent)
-- The sidebar shows an active loop indicator with cadence "1m"
-- The system schedules the next run for 1 minute from now
-- `/loop status` returns valid schedule data (cadence label, next run time, iteration count)
-- `/loop stop` deactivates the loop and clears the schedule
-
-### Verified
-- [ ] "Loop started" confirmation message appears
-- [ ] First run executes immediately
-- [ ] Sidebar shows 🔄 Active with cadence "1m"
-- [ ] `/loop status` shows schedule data
-- [ ] `/loop stop` stops the loop
-- [ ] `/loop status` returns null after stop
-
----
-
-## Test 3: /loop command — stop
-
-### Step 3 — Stop the loop
-Type the following in the CLI:
-```
-/loop stop
-```
-
-### Expected Result
-- The CLI shows a "Loop stopped" confirmation
-- The sidebar loop status changes to "No active loop"
-- The schedule is cleared (`/loop status` returns null)
-
-### Verified
-- [ ] "Loop stopped" confirmation appears
-- [ ] Sidebar shows "No active loop"
-- [ ] `/loop status` returns null
-
----
-
-## Test 4: Goal evaluation across loop runs
-
-### Step 4 — Start a loop with a goal
-Type the following in the CLI:
-```
-/goal all tests pass with no new failures
-```
-Then:
-```
-/loop 2m run the test suite and report results
-```
-
-### Expected Result
-- The goal condition persists across loop iterations
-- Each loop run receives the goal condition in context
-- The agent can evaluate whether the goal is satisfied after each run
-- When the goal is satisfied, the loop stops (or the agent reports goal completion)
-
-### Verified
-- [ ] Goal condition persists between loop iterations
-- [ ] Each loop run includes the goal condition in context
-- [ ] Agent reports goal status after each run
-
----
-
-## Test 5: Sidebar loop indicator
-
-### Step 5 — Verify sidebar integration
-While a loop is active:
-1. Check the right sidebar for the "Loop" section
-2. Verify it shows: Status (🔄 Active), Cadence, Next run, Iterations
-3. Stop the loop and verify the sidebar shows "No active loop"
-
-### Verified
-- [ ] Sidebar shows Loop section when active
-- [ ] Loop section shows correct cadence and next run time
-- [ ] Loop section shows iteration count
-- [ ] Loop section shows "No active loop" when stopped
-
----
-
-## Exit Criteria
-All 5 test steps pass with no errors. The goal/loop system is fully functional:
-1. `/goal` sets a condition that persists across loop runs
-2. `/loop` runs on a cadence schedule with recurring execution
-3. `/loop stop` deactivates the loop and clears state
-4. `/loop status` reflects accurate state
-5. The sidebar shows real-time loop status
-
-## Notes
-This test prompt is designed for manual CLI testing via tmux. Run each step
-sequentially and verify the expected result before proceeding to the next step.

package/dev/test-prompts/archive/rebrand-qa.md

@@ -1,258 +0,0 @@
-# Savant-Code Rebrand QA — Comprehensive A-Z Live Test
-
-**Purpose:** Final verification pass before public rebrand of SavantCode → Savant-Code. Covers every change from FID-001 through FID-016 plus end-to-end CLI/SDK/agent-runtime integration.
-
-**Mode:** Interactive live execution inside the Savant CLI. Drive the actual chat session, run real commands, exercise real tools. Report PASS/FAIL/SKIP with evidence for each item.
-
-**Environment:** `strict_mode: true`, `language: typescript`, `protocol.config.yaml` active, ECHO v0.2.0 loaded.
-
-**Skip Rules:**
-- Items marked `[LIVE]` require interactive CLI drives — execute in tmux.
-- Items marked `[SOURCE]` require code-verification only — read the file + grep.
-- Skip any item that fails on the first try — do not retry. Capture the exact error.
-
----
-
-## Section 1: Boot & Identity (FID-008)
-
-1. `[LIVE]` Drive `/help` → verify all 9 ECHO agents listed in `Ecosystem`
-2. `[SOURCE]` Read `ECHO.md` lines 1-50 → confirm "Savant" identity, v0.2.0, ECHO loaded
-3. `[LIVE]` Confirm `strict_mode: true` injected into system prompt (no `strict_mode=false` disclaimer)
-4. `[LIVE]` Confirm `language: typescript` present in agent boot
-5. `[LIVE]` Confirm ECHO_BOOT_CONTEXT shows open FIDs from `dev/fids/`
-6. `[LIVE]` Confirm session summary auto-created at `dev/session-summaries/<date>.md`
-7. `[LIVE]` Drive `/fids` → list all open FIDs with status (OPEN / ARCHIVED)
-
----
-
-## Section 2: Slash Commands (FID-012)
-
-8. `[LIVE]` Drive `/phase` → print current FSM phase
-9. `[LIVE]` Drive `/phase green` → transition illegal → expect "INVALID FSM transition: idle → green"
-10. `[LIVE]` Drive `/phase red` → transition legal (from idle) → expect "FSM phase: red"
-11. `[LIVE]` Drive `/fid FID-2026-0718-016` → show FID summary (was pre-existing test failures)
-12. `[LIVE]` Drive `/dev` → toggle dev mode (FID-003)
-13. `[LIVE]` With dev mode ON, drive a write_file from Orchestrator to `dev/scratchpad/test.md` → expect succeeds
-14. `[LIVE]` With dev mode OFF, drive same write_file → expect BLOCKED unless path is dev/scratchpad or dev/fids
-
----
-
-## Section 3: FSM Phase Tracking & SYNCHRONIZATION (FID-009 + FID-008 F10)
-
-15. `[LIVE]` Confirm right sidebar `phase` indicator matches runtime FSM (idle by default)
-16. `[LIVE]` Drive `/phase red` → confirm sidebar updates to `[RED]` (no refresh needed)
-17. `[LIVE]` Drive `/phase green` → confirm sidebar → `[GREEN]`
-18. `[LIVE]` Drive a write_file from Orchestrator during GREEN → expect succeeds to non-system path
-19. `[LIVE]` Drive `/phase audit` → confirm sidebar → `[AUDIT]`
-20. `[LIVE]` Drive `run_terminal_command` during AUDIT → expect succeeds
-21. `[LIVE]` Drive same tool during RED → expect BLOCKED with "INVALID FSM transition: terminal commands available during AUDIT"
-22. `[LIVE]` Drive `/phase complete` → confirm sidebar resets to `[IDLE]` after FID archive
-23. `[LIVE]` Confirm status banner above input also resets from "working" to idle (per FID-008 F10)
-24. `[LIVE]` Confirm `/phase idle` auto-fires on new user message after complete
-
----
-
-## Section 4: Agent Roster & Separation of Duties (FID-006)
-
-25. `[SOURCE]` Read `agents/orchestrator/base2.ts` → confirm tools: spawn_agents, read_files, transition_phase, write_todos, render_ui, etc. NO write_file for production paths (only via Forge). WRITE TO SCRATCHPAD OK (FID-008).
-26. `[SOURCE]` Read `agents/detective/detective.ts` → confirm tools: code_search, set_output ONLY. No write.
-27. `[SOURCE]` Read `agents/forge/forge.ts` → confirm tools: write_file, str_replace, set_output. No bash.
-28. `[SOURCE]` Read `agents/verifier/verifier.ts` → confirm NO write tools. Reads via message history.
-29. `[SOURCE]` Read `agents/recorder/recorder.ts` → confirm transition_phase tool available.
-30. `[SOURCE]` Read `agents/thinker/thinker.ts` → confirm sequentialthinking tool, no writes.
-31. `[SOURCE]` Read `agents/scout/scout.ts` → confirm read-only tools only.
-32. `[SOURCE]` Read `agents/researcher/researcher.ts` → confirm web_search, read_url.
-33. `[SOURCE]` Read `agents/scribe/scribe.ts` → confirm docs tools.
-
----
-
-## Section 5: Subagent FSM Inheritance (FID-004)
-
-34. `[SOURCE]` Read `packages/agent-runtime/src/tools/spawn-agent-utils.ts` `createAgentState()` → confirm fsmPhase + iterationCount inherited from parent.
-35. `[SOURCE]` Confirm subagents DO inherit strictMode, fsmPhase, iterationCount.
-36. `[LIVE]` Spawn Detective from Orchestrator (during RED) → confirm Detective's fsmPhase = RED (not idle).
-37. `[LIVE]` Spawn Forge during GREEN → confirm Forge can write (subagent fsmPhase = GREEN).
-
----
-
-## Section 6: Tool Gating & Strict Mode (FID-001 + FID-004)
-
-38. `[LIVE]` During RED, drive Orchestrator write_file to `src/example.ts` → expect BLOCKED.
-39. `[LIVE]` During GREEN, same write_file → expect succeeds.
-40. `[LIVE]` During RED, drive run_terminal_command `ls` → expect BLOCKED.
-41. `[LIVE]` During AUDIT, same → expect succeeds.
-42. `[LIVE]` Confirm `strict_mode: true` set globally (`protocol.config.yaml`).
-43. `[SOURCE]` Read `packages/agent-runtime/src/tools/tool-executor.ts` lines 339-362 → confirm gate logic.
-
----
-
-## Section 7: FID-Bound Execution & Snapshots (FID-001 + FID-005)
-
-44. `[LIVE]` Drive `/phase red` → confirm gate blocks write_file even if no FID open.
-45. `[LIVE]` Drive `/phase green` with no FID open → confirm blocked (hasOpenFids check).
-46. `[LIVE]` Create FID-2026-0718-017 manually, then drive `/phase green` → confirm succeeds.
-47. `[LIVE]` During GREEN, Forge writes `src/example.ts` → confirm snapshot stored at `dev/scratchpad/.snapshots/`.
-48. `[LIVE]` During self_correct → green (after audit fail), confirm original `src/example.ts` restored.
-49. `[LIVE]` After green passes, confirm file content matches new version.
-50. `[LIVE]` After complete, confirm snapshot cleared.
-
----
-
-## Section 8: Circuit Breakers (FID-007)
-
-51. `[LIVE]` Drive 11 iterations of self_correct → green → confirm hard stop at iteration 10.
-52. `[LIVE]` Drive 11 iterations of red → green → confirm same stop.
-53. `[SOURCE]` Read `packages/agent-runtime/src/tools/handlers/tool/transition-phase.ts` → confirm iterationCount enforcement.
-
----
-
-## Section 9: MCP Proxy Timeout & Fallback (FID-007)
-
-54. `[SOURCE]` Read `common/src/mcp/client.ts` → confirm withTimeout per-server (2s default).
-55. `[SOURCE]` Read for fallback logic when MCP server disables proxy.
-56. `[LIVE]` Drive `get_mcp_tools` against slow server → confirm timeout error surfaces.
-57. `[LIVE]` Drive same with proxy disabled → confirm fallback path works.
-
----
-
-## Section 10: Skills System (FID-002)
-
-58. `[SOURCE]` Read `.agents/skills/` → confirm all 7 SKILL.md present (typescript, python, rust, java, go, csharp, release-workflow).
-59. `[SOURCE]` Each SKILL.md has YAML frontmatter (name, description).
-60. `[LIVE]` Trigger codebase task with `.py` files → confirm Python skill loaded by agent.
-61. `[LIVE]` Trigger codebase task with `.ts` files → confirm TypeScript skill loaded.
-62. `[LIVE]` Confirm loaded skills list visible in right sidebar (if shown) or session context.
-
----
-
-## Section 11: Knowledge Files (FID-005)
-
-63. `[LIVE]` Confirm `LEARNINGS.md`, `CLAUDE.md`, `AGENTS.md` loaded into agent context.
-64. `[LIVE]` Confirm `dev/session-summaries/*.md` referenced in agent boot.
-65. `[LIVE]` Confirm `dev/fids/open/` listed in agent boot (open FIDs).
-66. `[LIVE]` Confirm KNOWLEDGE_FILE_NAMES filter works (no subdirectory injection).
-
----
-
-## Section 12: Path Normalization & Cross-Platform (FID-015 + FID-016)
-
-67. `[SOURCE]` Read `sdk/src/tools/path-utils.ts` → confirm toPosix helper.
-68. `[LIVE]` Run `bun test sdk/src/__tests__/user-knowledge-files.test.ts` → 15/15 pass.
-69. `[LIVE]` Run `bun test sdk/src/__tests__/code-search.test.ts -t 'cwd'` → 3/3 pass.
-70. `[LIVE]` Run `bun test sdk/src/__tests__/database.test.ts` → 3/3 pass.
-71. `[LIVE]` Run `bun test sdk/src/__tests__/initial-session-state.test.ts -t 'discovers'` → 1/1 pass.
-72. `[LIVE]` Run `bun test sdk/src/__tests__/load-agents.test.ts -t 'verbose'` → 1/1 pass.
-73. `[LIVE]` Run `bun test sdk/src/__tests__/load-skills.test.ts -t 'malformed'` → 1/1 pass.
-74. `[LIVE]` Run `bun test sdk/e2e/custom-agents/apply-patch-tool.e2e.test.ts` → 1/1 pass (or skip).
-75. `[LIVE]` Run `bun test sdk/src/**/*` full suite → 488/488 expected pass.
-
----
-
-## Section 13: Typecheck (FID-015 + FID-016 regression)
-
-76. `[LIVE]` `bun run --cwd=common typecheck` → exit 0, no errors.
-77. `[LIVE]` `bun run --cwd=agent-runtime typecheck` → exit 0.
-78. `[LIVE]` `bun run --cwd=cli typecheck` → exit 0.
-79. `[LIVE]` `bun run --cwd=sdk typecheck` → exit 0.
-
----
-
-## Section 14: Right Sidebar (FID-016 visual)
-
-80. `[LIVE]` Confirm "One Mind. A Thousand Faces." tagline visible (single line).
-81. `[LIVE]` Confirm TokenMeter shows inline (single row).
-82. `[LIVE]` Confirm Model name shows full (20 chars, not 14).
-83. `[LIVE]` Confirm Context section has own bordered sub-section.
-84. `[LIVE]` Confirm ECHO Protocol section above Session.
-85. `[LIVE]` Drive transition → confirm phase color updates (red=#ef4444, green=#39ff14, audit=#eab308).
-
----
-
-## Section 15: Status Banner & Working Indicator (FID-008 F10)
-
-86. `[LIVE]` During thinking → confirm banner shows "thinking..." (not just "working").
-87. `[LIVE]` During tool execution → confirm banner shows tool name.
-88. `[LIVE]` During subagent spawn → confirm banner shows agent name.
-89. `[LIVE]` After idle → confirm banner resets to ready/prompt.
-
----
-
-## Section 16: ECHO Roster Fidelity (ECHO.md ↔ agent runtime)
-
-90. `[SOURCE]` Confirm ECHO.md agent table matches all 9 directories in `agents/`.
-91. `[SOURCE]` Confirm savant-free-legacy agents removed (any in `agents/` not listed in ECHO.md is a violation).
-92. `[SOURCE]` Confirm 4 SAVANT_GEMINI_THINKER_* variants consolidated into Verifier (FID-006).
-
----
-
-## Section 17: CLI Edge Cases
-
-93. `[LIVE]` Drive `/exit` → confirm clean shutdown, session summary saved.
-94. `[LIVE]` Drive empty message → confirm ignored (no error).
-95. `[LIVE]` Drive very long message → confirm streamed back.
-96. `[LIVE]` Drive message with markdown → confirm rendered correctly.
-97. `[LIVE]` Drive multi-line agent output → confirm pagination/scroll works.
-
----
-
-## Section 18: Database & Persistence (FID-005)
-
-98. `[SOURCE]` Read `packages/database/` → confirm `~/.savant/data.db` path (not savant-free).
-99. `[LIVE]` Confirm session state persists across restarts.
-100. `[LIVE]` Confirm cost/tokens tracked across sessions in DB.
-
----
-
-## Section 19: Bundled Agents (FID-007 + FID-006 sync)
-
-101. `[SOURCE]` Read `packages/agent-runtime/src/tools/handlers/tool/bundled-agents.generated.ts` → confirm Scout + Detective + Verifier + Forge + Recorder + Scribe all present.
-102. `[SOURCE]` Confirm Scout + MCP timeout changes reflected (rebuild if needed).
-103. `[LIVE]` Drive spawn_agents with each type → confirm all return valid results.
-
----
-
-## Section 20: End-to-End Coding Flow
-
-104. `[LIVE]` Start session, describe simple feature → confirm Orchestrator creates FID.
-105. `[LIVE]` FID goes RED → GREEN → AUDIT → COMPLETE → confirm all phases visible.
-106. `[LIVE]` Confirm file Git diff shows only approved changes.
-107. `[LIVE]` Drive tool failure (e.g., invalid path) → confirm error caught + agent retries or self-corrects.
-108. `[LIVE]` Drive oscillation (same error 3x) → confirm circuit breaker detects.
-
----
-
-## Section 21: Free-Buff Mode Considerations (future)
-
-109. `[SOURCE]` Confirm IS_SAVANT_FREE constant exists.
-110. `[SOURCE]` Confirm savant-free-model-store.ts present (uses model selection).
-111. `[SOURCE]` Confirm ad-supported flow paths marked but not blocking paid free flow.
-
----
-
-## Reporting Format
-
-For each item, output:
-```
-### Item N: [SHORT_TITLE]
-**Status:** PASS / FAIL / SKIP
-**Mode:** [LIVE] / [SOURCE]
-**Evidence:** [file path, line number, or command output]
-**Error:** [exact error message if FAIL]
-**Timestamp:** [HH:MM]
-```
-
-End with summary table: passes count, fails count, skips count, blockers list.
-
----
-
-## Pre-Rebrand Blockers
-
-Any FAIL in Sections 3 (FSM sync), 6 (tool gating), 8 (circuit breakers), 12-13 (tests/typecheck), or 15 (banner sync) IS A BLOCKER. Fix before rebrand.
-
-All other FAILs are release candidates - triage and ship with known issues list.
-
----
-
-**Run by:** Savant Orchestrator (you)
-**Audited by:** Nova (external ECHO compliance review)
-**Approved by:** Creator (user)