chore: remove llm integration tests and prep release docs

10 files changed, 0 insertions, 530 deletions

diff --git a/tests/integration/ext/herdr-test.ts b/tests/integration/ext/herdr-test.ts
deleted file mode 100644
index 4b6c17d5..00000000
--- a/tests/integration/ext/herdr-test.ts
+++ /dev/null
@@ -1,146 +0,0 @@
-/**
- * Minimal tmux extension for herdr integration testing.
- *
- * Two actions: send and read. No wait, no pane creation, no discovery loops.
- * Talks directly to a pane tagged @pi_name=herdr.
- */
-
-import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
-import { StringEnum } from "@mariozechner/pi-ai";
-import { Text } from "@mariozechner/pi-tui";
-import { Type } from "@sinclair/typebox";
-
-function stripAnsi(text: string): string {
-	return text
-		.replace(/\x1b\].*?(?:\x07|\x1b\\)/g, "")
-		.replace(/\x1bPtmux;.*?\x1b\\/g, "")
-		.replace(/\x1b\[[0-9;]*[a-zA-Z]/g, "")
-		.replace(/\x1b[^[\]P]/g, "")
-		.replace(/\r/g, "");
-}
-
-export default function (pi: ExtensionAPI) {
-	let herdrPaneId: string | null = null;
-	let windowId: string | null = null;
-
-	pi.on("session_start", async () => {
-		if (!process.env.TMUX) return;
-
-		try {
-			const winResult = await pi.exec("tmux", [
-				"display-message", "-p", "-t", process.env.TMUX_PANE || "",
-				"#{window_id}",
-			]);
-			if (winResult.code === 0) windowId = winResult.stdout.trim();
-
-			if (windowId) {
-				const listResult = await pi.exec("tmux", [
-					"list-panes", "-t", windowId,
-					"-F", "#{pane_id}\t#{@pi_name}",
-				]);
-				if (listResult.code === 0) {
-					for (const line of listResult.stdout.trim().split("\n")) {
-						const [paneId, name] = line.split("\t");
-						if (name === "herdr") {
-							herdrPaneId = paneId;
-							break;
-						}
-					}
-				}
-			}
-		} catch {}
-	});
-
-	pi.registerTool({
-		name: "herdr",
-		label: "herdr",
-		description:
-			"Interact with the herdr instance under test. " +
-			"Actions: send (send keys or text to herdr), read (capture herdr screen).",
-		parameters: Type.Object({
-			action: StringEnum(["send", "read"] as const, {
-				description: "Action: send or read",
-			}),
-			keys: Type.Optional(
-				Type.String({
-					description:
-						"Keys to send, space-separated (for send). Examples: C-s, Enter, n, v, Escape",
-				}),
-			),
-			text: Type.Optional(
-				Type.String({ description: "Literal text to type (for send). Sent as-is." }),
-			),
-			lines: Type.Optional(
-				Type.Number({ description: "Scrollback lines to capture (for read, default: 50)" }),
-			),
-		}),
-
-		async execute(_toolCallId, params) {
-			if (!herdrPaneId) {
-				throw new Error("herdr pane not found. Is herdr running in this tmux window?");
-			}
-
-			switch (params.action) {
-				case "send": {
-					if (!params.keys && !params.text) {
-						throw new Error("send requires 'keys' or 'text'");
-					}
-					if (params.text) {
-						await pi.exec("tmux", [
-							"send-keys", "-l", "-t", herdrPaneId, params.text,
-						]);
-					}
-					if (params.keys) {
-						const keyArgs = params.keys.split(/\s+/).filter(Boolean);
-						await pi.exec("tmux", [
-							"send-keys", "-t", herdrPaneId, ...keyArgs,
-						]);
-					}
-					const desc = [
-						params.text && `"${params.text}"`,
-						params.keys,
-					].filter(Boolean).join(" + ");
-					return {
-						content: [{ type: "text", text: `Sent: ${desc}` }],
-					};
-				}
-
-				case "read": {
-					const lines = params.lines ?? 50;
-					const result = await pi.exec("tmux", [
-						"capture-pane", "-t", herdrPaneId, "-p", "-S", `-${lines}`,
-					]);
-					if (result.code !== 0) {
-						throw new Error(`capture-pane failed: ${result.stderr}`);
-					}
-					let output = stripAnsi(result.stdout);
-					output = output.replace(/\n+$/, "\n");
-					return {
-						content: [{ type: "text", text: output }],
-					};
-				}
-
-				default:
-					throw new Error(`Unknown action: ${params.action}`);
-			}
-		},
-
-		renderCall(args, theme) {
-			const action = args.action || "?";
-			let t = theme.fg("toolTitle", theme.bold("herdr "));
-			t += theme.fg("accent", action);
-			if (args.text) t += theme.fg("dim", ` "${args.text}"`);
-			if (args.keys) t += theme.fg("dim", ` ${args.keys}`);
-			return new Text(t, 0, 0);
-		},
-
-		renderResult(result, _opts, theme) {
-			const c = result.content?.[0];
-			const text = c?.type === "text" ? c.text : "";
-			if (text.length > 200) {
-				return new Text(theme.fg("dim", text.slice(0, 200) + "…"), 0, 0);
-			}
-			return new Text(theme.fg("dim", text), 0, 0);
-		},
-	});
-}
diff --git a/tests/integration/pdeathsig.py b/tests/integration/pdeathsig.py
deleted file mode 100755
index d4a25ac7..00000000
--- a/tests/integration/pdeathsig.py
+++ /dev/null
@@ -1,44 +0,0 @@
-#!/usr/bin/env python3
-"""Exec a command with PR_SET_PDEATHSIG — receive a signal when parent dies.
-
-Usage: pdeathsig.py SIGNAL COMMAND [ARGS...]
-
-Fallback for systems without `setpriv --pdeathsig`.
-"""
-import ctypes
-import ctypes.util
-import os
-import signal
-import sys
-
-PR_SET_PDEATHSIG = 1
-
-SIGNALS = {
-    "TERM": signal.SIGTERM, "SIGTERM": signal.SIGTERM,
-    "INT": signal.SIGINT, "SIGINT": signal.SIGINT,
-    "HUP": signal.SIGHUP, "SIGHUP": signal.SIGHUP,
-    "KILL": signal.SIGKILL, "SIGKILL": signal.SIGKILL,
-}
-
-def main() -> int:
-    if len(sys.argv) < 3:
-        print(f"usage: {sys.argv[0]} SIGNAL COMMAND [ARGS...]", file=sys.stderr)
-        return 2
-    sig_name = sys.argv[1]
-    argv = sys.argv[2:]
-    if sig_name not in SIGNALS:
-        print(f"unknown signal: {sig_name}", file=sys.stderr)
-        return 2
-    sig = SIGNALS[sig_name]
-    libc = ctypes.CDLL(ctypes.util.find_library("c") or "libc.so.6", use_errno=True)
-    if libc.prctl(PR_SET_PDEATHSIG, sig, 0, 0, 0) != 0:
-        err = ctypes.get_errno()
-        raise OSError(err, os.strerror(err))
-    # Race fix: parent may have died before prctl took effect
-    if os.getppid() == 1:
-        os.kill(os.getpid(), sig)
-    os.execvp(argv[0], argv)
-    return 127
-
-if __name__ == "__main__":
-    raise SystemExit(main())
diff --git a/tests/integration/results/.gitignore b/tests/integration/results/.gitignore
deleted file mode 100644
index d6b7ef32..00000000
--- a/tests/integration/results/.gitignore
+++ /dev/null
@@ -1,2 +0,0 @@
-*
-!.gitignore
diff --git a/tests/integration/run_all.sh b/tests/integration/run_all.sh
deleted file mode 100755
index cd373af4..00000000
--- a/tests/integration/run_all.sh
+++ /dev/null
@@ -1,116 +0,0 @@
-#!/usr/bin/env bash
-#
-# Run all herdr integration tests in parallel.
-#
-# Each spec gets its own tmux server (isolated socket).
-# Workers are launched with PDEATHSIG so they get SIGTERM if this script dies.
-# A stale sweep cleans up sockets from any previous crashed run.
-#
-set -uo pipefail
-
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-SPECS_DIR="$SCRIPT_DIR/specs"
-RESULTS_DIR="$SCRIPT_DIR/results"
-TMUX_ROOT="${TMPDIR:-/tmp}/herdr-test-sockets"
-
-mkdir -p "$RESULTS_DIR" "$TMUX_ROOT"
-
-# Clean previous results
-rm -f "$RESULTS_DIR"/*.json "$RESULTS_DIR"/*.txt 2>/dev/null
-
-# Sweep stale tmux sockets from crashed previous runs
-for dir in "$TMUX_ROOT"/*/; do
-    [ -d "$dir" ] || continue
-    owner_pid=""
-    [ -f "$dir/owner.pid" ] && owner_pid="$(<"$dir/owner.pid")"
-    # If owning worker is still alive, leave it
-    if [ -n "$owner_pid" ] && kill -0 "$owner_pid" 2>/dev/null; then
-        continue
-    fi
-    # Dead owner — kill the tmux server and remove the dir
-    [ -S "$dir/tmux.sock" ] && tmux -S "$dir/tmux.sock" kill-server 2>/dev/null || true
-    rm -rf "$dir"
-done
-
-# Collect specs
-SPECS=("$SPECS_DIR"/*.md)
-if [ ${#SPECS[@]} -eq 0 ]; then
-    echo "No test specs found in $SPECS_DIR"
-    exit 0
-fi
-
-echo "Running ${#SPECS[@]} integration tests..."
-
-# Detect PDEATHSIG support
-PDEATHSIG_CMD=()
-if command -v setpriv >/dev/null 2>&1 && setpriv --help 2>&1 | grep -q -- '--pdeathsig'; then
-    PDEATHSIG_CMD=(setpriv --pdeathsig TERM --)
-elif [ -f "$SCRIPT_DIR/pdeathsig.py" ]; then
-    PDEATHSIG_CMD=(python3 "$SCRIPT_DIR/pdeathsig.py" TERM)
-fi
-
-# Launch all specs in parallel — directly from this shell (no subshell wrapper)
-PIDS=()
-NAMES=()
-
-for spec in "${SPECS[@]}"; do
-    name="$(basename "$spec" .md)"
-    NAMES+=("$name")
-
-    "${PDEATHSIG_CMD[@]}" "$SCRIPT_DIR/run_spec.sh" "$spec" "$RESULTS_DIR" "$TMUX_ROOT" &
-    PIDS+=($!)
-done
-
-# Wait for all, ignoring individual exit codes (we read results from files)
-for pid in "${PIDS[@]}"; do
-    wait "$pid" 2>/dev/null || true
-done
-
-echo ""
-
-# Print summary from result files
-PASSED=0
-FAILED=0
-SKIPPED=0
-
-for name in "${NAMES[@]}"; do
-    result_file="$RESULTS_DIR/$name.json"
-
-    if [[ ! -f "$result_file" ]]; then
-        echo "  ? $name (no result file)"
-        FAILED=$((FAILED + 1))
-        continue
-    fi
-
-    result=$(python3 -c "import sys,json; print(json.load(sys.stdin).get('result','unknown'))" < "$result_file" 2>/dev/null || echo "unknown")
-
-    case "$result" in
-        pass)
-            echo "  ✓ $name"
-            PASSED=$((PASSED + 1))
-            ;;
-        skip)
-            echo "  - $name (skipped)"
-            SKIPPED=$((SKIPPED + 1))
-            ;;
-        *)
-            echo "  ✗ $name"
-            python3 -c "
-import sys, json
-data = json.load(sys.stdin)
-for c in data.get('checks', []):
-    if not c.get('pass', True):
-        print(f'      ✗ {c[\"name\"]}: {c.get(\"detail\", \"\")}')
-notes = data.get('notes', '')
-if notes and data.get('result') == 'error':
-    print(f'      {notes}')
-" < "$result_file" 2>/dev/null
-            FAILED=$((FAILED + 1))
-            ;;
-    esac
-done
-
-echo ""
-echo "$PASSED passed, $FAILED failed, $SKIPPED skipped"
-
-[ $FAILED -eq 0 ]
diff --git a/tests/integration/run_spec.sh b/tests/integration/run_spec.sh
deleted file mode 100755
index 43056f6e..00000000
--- a/tests/integration/run_spec.sh
+++ /dev/null
@@ -1,126 +0,0 @@
-#!/usr/bin/env bash
-#
-# Run a single herdr integration test spec via pi.
-#
-# Usage: ./run_spec.sh <spec.md> [results_dir] [tmux_socket_root]
-#
-# Each invocation creates its own tmux server (isolated socket).
-# Cleanup kills the entire server — no orphaned sessions possible.
-#
-set -euo pipefail
-
-SPEC="$1"
-SPEC_NAME="$(basename "$SPEC" .md)"
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-PROJECT_DIR="$(cd "$SCRIPT_DIR/../.." && pwd)"
-RESULTS_DIR="${2:-$SCRIPT_DIR/results}"
-TMUX_ROOT="${3:-${TMPDIR:-/tmp}/herdr-test-sockets}"
-SYSTEM_PROMPT_FILE="$SCRIPT_DIR/system.md"
-HERDR="$PROJECT_DIR/target/release/herdr"
-EXTENSION="$SCRIPT_DIR/ext/herdr-test.ts"
-MODEL="${HERDR_TEST_MODEL:-minimax/MiniMax-M2.7}"
-
-mkdir -p "$RESULTS_DIR" "$TMUX_ROOT"
-
-# Per-spec tmux socket — this spec owns its entire tmux server
-TMUX_DIR="$(mktemp -d "$TMUX_ROOT/${SPEC_NAME}.XXXXXX")"
-TMUX_SOCKET="$TMUX_DIR/tmux.sock"
-SESSION_NAME="test"
-
-# Track ownership for stale sweep
-printf '%s\n' "$$" > "$TMUX_DIR/owner.pid"
-
-# Temp files (initialized for safe cleanup)
-PROMPT_FILE=""
-RAW_OUTPUT=""
-
-cleanup() {
-    # Kill our entire tmux server — takes everything with it
-    tmux -S "$TMUX_SOCKET" kill-server 2>/dev/null || true
-    [[ -n "$PROMPT_FILE" ]] && rm -f "$PROMPT_FILE" 2>/dev/null || true
-    rm -rf "$TMUX_DIR" 2>/dev/null || true
-    # RAW_OUTPUT is moved to results dir, not cleaned here
-}
-trap cleanup EXIT INT TERM HUP
-
-TMUX=(tmux -S "$TMUX_SOCKET")
-
-# Prerequisites
-RESULT_FILE="$RESULTS_DIR/$SPEC_NAME.json"
-if ! command -v tmux &>/dev/null; then echo '{"test":"'"$SPEC_NAME"'","result":"skip","checks":[],"notes":"tmux not found"}' > "$RESULT_FILE"; exit 0; fi
-if ! command -v pi &>/dev/null; then echo '{"test":"'"$SPEC_NAME"'","result":"skip","checks":[],"notes":"pi not found"}' > "$RESULT_FILE"; exit 0; fi
-if [[ ! -x "$HERDR" ]]; then echo '{"test":"'"$SPEC_NAME"'","result":"error","checks":[],"notes":"herdr binary not found"}' > "$RESULT_FILE"; exit 1; fi
-
-# Start herdr in its own tmux server
-"${TMUX[@]}" new-session -d -s "$SESSION_NAME" -x 120 -y 50
-"${TMUX[@]}" send-keys -t "$SESSION_NAME" "$HERDR --no-session" Enter
-HERDR_PANE=$("${TMUX[@]}" list-panes -t "$SESSION_NAME" -F '#{pane_id}' | head -1)
-"${TMUX[@]}" set-option -p -t "$HERDR_PANE" @pi_name "herdr"
-
-sleep 1
-
-# Build prompt
-PROMPT_FILE=$(mktemp /tmp/herdr-test-prompt-XXXXX.md)
-RAW_OUTPUT=$(mktemp /tmp/herdr-test-raw-XXXXX.txt)
-cat "$SYSTEM_PROMPT_FILE" > "$PROMPT_FILE"
-echo -e "\n---\n" >> "$PROMPT_FILE"
-cat "$SPEC" >> "$PROMPT_FILE"
-
-# Run pi in the same tmux server
-SESSION_DIR="$RESULTS_DIR/sessions"
-mkdir -p "$SESSION_DIR"
-
-"${TMUX[@]}" split-window -t "$SESSION_NAME" -h \
-    "cd $PROJECT_DIR && pi -p \
-        --session-dir '$SESSION_DIR' \
-        --no-extensions --no-skills --no-prompt-templates \
-        -e '$EXTENSION' \
-        --no-tools \
-        --model '$MODEL' \
-        --system-prompt '$PROMPT_FILE' \
-        'Execute the test now. Use the herdr tool to send keys and read the screen. Output the JSON result block when done.' \
-        > '$RAW_OUTPUT' 2>&1; \
-     echo '___DONE___' >> '$RAW_OUTPUT'; \
-     sleep 1"
-
-# Wait for completion
-TIMEOUT=180
-ELAPSED=0
-while [ $ELAPSED -lt $TIMEOUT ]; do
-    if grep -q '___DONE___' "$RAW_OUTPUT" 2>/dev/null; then
-        break
-    fi
-    sleep 2
-    ELAPSED=$((ELAPSED + 2))
-done
-
-if [ $ELAPSED -ge $TIMEOUT ]; then
-    echo '{"test":"'"$SPEC_NAME"'","result":"error","checks":[],"notes":"pi timed out after '"$TIMEOUT"'s"}' > "$RESULT_FILE"
-    cp "$RAW_OUTPUT" "$RESULTS_DIR/$SPEC_NAME.raw.txt" 2>/dev/null || true
-    rm -f "$RAW_OUTPUT"
-    exit 1
-fi
-
-# Always save raw output
-cp "$RAW_OUTPUT" "$RESULTS_DIR/$SPEC_NAME.raw.txt" 2>/dev/null || true
-rm -f "$RAW_OUTPUT"
-
-# Extract JSON
-JSON_RESULT=$(awk '/^```json$/,/^```$/' "$RESULTS_DIR/$SPEC_NAME.raw.txt" | grep -v '^```')
-
-if [[ -z "$JSON_RESULT" ]]; then
-    echo '{"test":"'"$SPEC_NAME"'","result":"error","checks":[],"notes":"no JSON in pi output"}' > "$RESULT_FILE"
-    exit 1
-fi
-
-if echo "$JSON_RESULT" | python3 -m json.tool > "$RESULT_FILE" 2>/dev/null; then
-    RESULT=$(python3 -c "import sys,json; print(json.load(sys.stdin).get('result','unknown'))" < "$RESULT_FILE")
-    if [[ "$RESULT" == "pass" ]]; then
-        exit 0
-    else
-        exit 1
-    fi
-else
-    echo '{"test":"'"$SPEC_NAME"'","result":"error","checks":[],"notes":"invalid JSON from pi"}' > "$RESULT_FILE"
-    exit 1
-fi
diff --git a/tests/integration/specs/create_workspace.md b/tests/integration/specs/create_workspace.md
deleted file mode 100644
index 0f588c69..00000000
--- a/tests/integration/specs/create_workspace.md
+++ /dev/null
@@ -1,12 +0,0 @@
-# Test: Create workspace
-
-## Steps
-
-1. Read the screen to verify herdr started in empty state (should show "No active workspace" or "Press n to create one")
-2. Press `n` to enter create mode
-3. Type `test-workspace`
-4. Press Enter to create
-5. Read the screen and verify:
-   - The sidebar shows `1` followed by `test-workspace`
-   - A shell prompt is visible in the main pane area
-   - The active marker `▸` is next to the workspace
diff --git a/tests/integration/specs/sidebar_collapse.md b/tests/integration/specs/sidebar_collapse.md
deleted file mode 100644
index b5d42c25..00000000
--- a/tests/integration/specs/sidebar_collapse.md
+++ /dev/null
@@ -1,13 +0,0 @@
-# Test: Sidebar collapse and expand
-
-## Steps
-
-1. Press `n` to create a workspace named `test`
-2. Read the screen and verify the expanded sidebar shows the full workspace name `test`
-3. Press prefix (ctrl+s), then `b` to collapse the sidebar
-4. Read the screen and verify:
-   - The sidebar is now narrow (just numbers and state dots, no workspace name visible)
-   - The `»` expand icon is visible at the bottom of the sidebar
-   - The terminal pane takes up more horizontal space
-5. Press prefix (ctrl+s), then `b` again to expand
-6. Verify the full workspace name `test` is visible again in the sidebar
diff --git a/tests/integration/specs/split_and_navigate.md b/tests/integration/specs/split_and_navigate.md
deleted file mode 100644
index cb75d72d..00000000
--- a/tests/integration/specs/split_and_navigate.md
+++ /dev/null
@@ -1,11 +0,0 @@
-# Test: Split pane and navigate
-
-## Steps
-
-1. Press `n` to create a workspace named `dev`
-2. Press prefix (ctrl+s), then `v` to split vertically
-3. Read the screen and verify two panes are visible side by side, each with a shell prompt
-4. Verify the sidebar shows a pane count indicator like `[2]`
-5. Press prefix (ctrl+s), then `h` to navigate to the left pane
-6. Press prefix (ctrl+s), then `l` to navigate back to the right pane
-7. Verify the thick border (focused indicator) moved between panes
diff --git a/tests/integration/specs/workspace_switching.md b/tests/integration/specs/workspace_switching.md
deleted file mode 100644
index 0711c575..00000000
--- a/tests/integration/specs/workspace_switching.md
+++ /dev/null
@@ -1,12 +0,0 @@
-# Test: Workspace switching with prefix+N
-
-## Steps
-
-1. Press `n` to create a workspace named `first`
-2. Press prefix (ctrl+s), then `n` to create a second workspace named `second`
-3. Read the screen and verify the sidebar shows two workspaces numbered 1 and 2
-4. Verify `second` is the active workspace (has the `▸` marker)
-5. Press prefix (ctrl+s), then `1` to switch to the first workspace
-6. Read the screen and verify `first` is now active (has the `▸` marker)
-7. Press prefix (ctrl+s), then `2` to switch back to `second`
-8. Verify `second` is active again
diff --git a/tests/integration/system.md b/tests/integration/system.md
deleted file mode 100644
index d6ccb05b..00000000
--- a/tests/integration/system.md
+++ /dev/null
@@ -1,48 +0,0 @@
-# herdr Integration Test Agent
-
-You are a test executor for herdr, a terminal workspace manager. Your job is to execute a test specification by interacting with herdr through tmux, then report whether the test passed or failed.
-
-## Environment
-
-- herdr is running in tmux pane `herdr` with `--no-session` (clean state, no saved workspaces)
-- You have the `tmux` tool to interact with it: `send` keys, `read` screen output
-- herdr starts in Navigate mode with no workspaces (empty state)
-
-## herdr Controls
-
-- **Navigate mode**: press prefix key (ctrl+s) from Terminal mode, or start with no workspaces
-- **Create workspace**: press `n`, type name, press Enter
-- **Switch workspace**: press `1`-`9` (by number) or arrow keys + Enter
-- **Split pane**: press `v` (vertical split) or `-` (horizontal split)
-- **Close pane**: press `x`
-- **Close workspace**: press `d` (then `y` to confirm)
-- **Collapse sidebar**: press `b`
-- **Resize mode**: press `r`, then h/j/k/l to resize, Esc to exit
-- **Back to terminal**: press Esc or prefix key again
-- **Quit**: press `q` in Navigate mode
-
-After creating a workspace or pressing Enter on one, you enter Terminal mode. Press ctrl+s (prefix) to go back to Navigate mode.
-
-## How to Execute
-
-1. Read the test specification carefully
-2. Execute each step using tmux send/read
-3. After each action, wait briefly then read the screen to verify
-4. When done, output your result as a JSON object
-
-## Output Format
-
-Your final message MUST be a JSON code block with this structure:
-
-```json
-{
-  "test": "<test name>",
-  "result": "pass" | "fail",
-  "checks": [
-    {"name": "<what was checked>", "pass": true|false, "detail": "<what you saw>"}
-  ],
-  "notes": "<any additional observations>"
-}
-```
-
-Be precise. Read the screen carefully. Don't assume — verify by reading.