@lossless-claude/lcm 0.2.0

package/.claude-plugin/skills/lcm-dogfood/SKILL.md

@@ -0,0 +1,102 @@
+---
+name: lcm-dogfood
+description: "This skill should be used when the user asks to \"run lcm dogfood\", \"self-test lcm\", \"validate lcm build\", \"test lcm hooks\", \"check lcm mcp tools\", or mentions lcm QA, regression testing, or pre-release validation. Runs a comprehensive 39-check self-test across all CLI commands, hooks, MCP tools, and resilience."
+user_invocable: true
+argument: "[phase]"
+---
+
+# lcm Dogfood — Live Self-Test Suite
+
+Comprehensive self-test for the lcm system in a live Claude Code session. Covers all 9 CLI commands, 4 hooks, 8 MCP tools, and resilience scenarios across 39 checks in 10 phases.
+
+**Arguments:** `all` (default), `health`, `import`, `compact`, `promote`, `sensitive`, `pipeline`, `hooks`, `mcp`, `resilience`, `debug`
+
+## Procedure
+
+Execute each phase in order (or just the requested phase). For each check:
+
+1. Run the command or verify the condition
+2. Record: ✅ PASS, ❌ FAIL, or ⚠️ SKIP (with reason)
+3. On FAIL: capture error, check daemon logs (`~/.lossless-claude/daemon.log`), continue
+4. Produce the **Scorecard** at the end
+5. Write failures to `.xgh/reviews/dogfood-YYYY-MM-DD.md`
+
+**Routing:** Use `ctx_execute` (context-mode sandbox) for commands producing large output. Use Bash for short-output commands. Use MCP tools directly for Phase 8.
+
+Consult `references/checks.md` for detailed pass/fail criteria for each check.
+
+## Phase Overview
+
+| # | Phase | Checks | What it tests |
+|---|-------|--------|---------------|
+| 1 | Health | 3 | Daemon status, doctor, version |
+| 2 | Import | 3 | Transcript ingestion + idempotency |
+| 3 | Compact | 3 | Summarization + idempotency |
+| 4 | Promote | 2 | Insight extraction + stats consistency |
+| 5 | Sensitive | 5 | Pattern list/test/add/remove cycle |
+| 6 | Pipeline | 2 | Full curate + diagnose |
+| 7 | Hooks | 6 | Wiring verification + live tests |
+| 8 | MCP | 8 | All 7 MCP tools + store-retrieve roundtrip |
+| 9 | Resilience | 3 | Kill/restart/graceful degradation |
+| 10 | Debug | 4 | Logs, PWD, DB existence, integrity |
+
+## Key Commands
+
+All CLI checks use `node dist/bin/lcm.js <subcommand>`. If `lcm` is on PATH, use that instead.
+
+### Hook Verification
+
+Hooks are registered in `.claude-plugin/plugin.json`, NOT `~/.claude/settings.json`. Verify all 4:
+- `SessionStart` → `lcm restore`
+- `UserPromptSubmit` → `lcm user-prompt`
+- `PreCompact` → `lcm compact`
+- `SessionEnd` → `lcm session-end`
+
+For live hook testing, pipe JSON to stdin:
+```bash
+echo '{}' | node dist/bin/lcm.js restore
+```
+
+The UserPromptSubmit hook requires `prompt` and `cwd` fields:
+```bash
+node -e 'console.log(JSON.stringify({prompt:"test query",cwd:process.cwd()}))' | node dist/bin/lcm.js user-prompt
+```
+
+### MCP Tool Testing
+
+Call lcm MCP tools directly from the session. All 8 tools to test:
+`lcm_doctor`, `lcm_stats`, `lcm_search`, `lcm_grep`, `lcm_store`, `lcm_expand`, `lcm_describe` + store-retrieve roundtrip.
+
+## Scorecard Template
+
+```
+| Phase       | Checks | ✅ Pass | ❌ Fail | ⚠️ Skip/Known |
+|-------------|--------|---------|---------|---------------|
+| Health      | 3      |         |         |               |
+| Import      | 3      |         |         |               |
+| Compact     | 3      |         |         |               |
+| Promote     | 2      |         |         |               |
+| Sensitive   | 5      |         |         |               |
+| Pipeline    | 2      |         |         |               |
+| Hooks       | 6      |         |         |               |
+| MCP         | 8      |         |         |               |
+| Resilience  | 3      |         |         |               |
+| Debug       | 4      |         |         |               |
+| **TOTAL**   | **39** |         |         |               |
+```
+
+For ❌ FAIL items, include: error message, daemon log excerpt, suggested fix.
+For ⚠️ KNOWN items, reference the bug number from `references/known-issues.md`.
+
+## Bundled Resources
+
+### Scripts
+
+Utility scripts for checks that require custom logic:
+- **`scripts/prompt-search-test.js`** — Test the daemon `/prompt-search` endpoint directly. Usage: `node scripts/prompt-search-test.js "query" [cwd]`
+- **`scripts/db-integrity.js`** — Check PRAGMA integrity_check on all project databases. Usage: `node scripts/db-integrity.js`
+
+### Reference Files
+
+- **`references/checks.md`** — All 39 checks with detailed pass/fail criteria, organized by phase
+- **`references/known-issues.md`** — Known bugs with root causes, affected checks, and fix status

package/.claude-plugin/skills/lcm-dogfood/references/checks.md

@@ -0,0 +1,239 @@
+# lcm Dogfood — Detailed Check Reference
+
+All 39 checks organized by phase. Each check includes the command to run and pass/fail criteria.
+
+## Phase 1: Health — 3 checks
+
+### 1.1 Daemon status
+```bash
+node dist/bin/lcm.js status
+```
+Pass if: daemon shows "up", project registered for current cwd, port 3737.
+
+### 1.2 Doctor
+```bash
+node dist/bin/lcm.js doctor
+```
+Pass if: all checks pass (8 passed, 0 failed). Record any warnings.
+
+### 1.3 Version
+```bash
+node dist/bin/lcm.js --version
+```
+Pass if: prints version string matching package.json.
+
+## Phase 2: Import — 3 checks
+
+### 2.1 Import session transcripts
+```bash
+node dist/bin/lcm.js import --all --verbose
+```
+Pass if: message count > 0, no errors. Use `ctx_execute` — output can be large.
+
+### 2.2 Status after import
+```bash
+node dist/bin/lcm.js status
+```
+Pass if: messages > 0, last ingest timestamp updated.
+
+### 2.3 Idempotent re-import
+Run import again immediately.
+Pass if: message count unchanged or only current-session messages added (small delta).
+
+## Phase 3: Compact — 3 checks
+
+### 3.1 Compact messages
+```bash
+node dist/bin/lcm.js compact --all --verbose
+```
+Pass if: summaries created, compression ratio reported. Note: calls LLM summarizer, may take minutes. Use 5-minute timeout.
+
+### 3.2 Status after compact
+```bash
+node dist/bin/lcm.js status
+```
+Pass if: summaries > 0.
+
+### 3.3 Idempotent re-compact
+Run compact again immediately.
+Pass if: no new summaries created.
+
+## Phase 4: Promote — 2 checks
+
+### 4.1 Promote insights
+```bash
+node dist/bin/lcm.js promote --all --verbose
+```
+Pass if: promoted count increments OR "no promotable content" (valid if summaries too short).
+
+### 4.2 Stats
+```bash
+node dist/bin/lcm.js stats --verbose
+```
+Pass if: shows messages, summaries, promoted counts, compression ratios. Numbers consistent with previous checks.
+
+## Phase 5: Sensitive Patterns — 5 checks
+
+### 5.1 List patterns
+```bash
+node dist/bin/lcm.js sensitive list
+```
+Pass if: shows 7 built-in patterns.
+
+### 5.2 Test scrubbing
+```bash
+node dist/bin/lcm.js sensitive test "my api key is sk-1234567890abcdefghij and password=hunter2"
+```
+Pass if: both API key and password are `[REDACTED]`.
+
+### 5.3 Add custom pattern
+```bash
+node dist/bin/lcm.js sensitive add "DOGFOOD_SECRET_\w+"
+```
+Pass if: pattern added.
+
+### 5.4 Test custom pattern
+```bash
+node dist/bin/lcm.js sensitive test "the value is DOGFOOD_SECRET_abc123"
+```
+Pass if: `DOGFOOD_SECRET_abc123` is `[REDACTED]`.
+
+### 5.5 Remove custom pattern
+```bash
+node dist/bin/lcm.js sensitive remove "DOGFOOD_SECRET_\w+"
+```
+Pass if: pattern removed. Run `sensitive list` to confirm.
+
+## Phase 6: Full Pipeline — 2 checks
+
+### 6.1 Curate (import + compact + promote)
+```bash
+node dist/bin/lcm.js import --all && node dist/bin/lcm.js compact --all && node dist/bin/lcm.js promote --all
+```
+Pass if: all three stages complete without error.
+
+### 6.2 Diagnose
+```bash
+node dist/bin/lcm.js diagnose --verbose
+```
+Pass if: no hook failures or ingestion gaps detected.
+
+## Phase 7: Hook Verification — 6 checks
+
+Hooks are registered in `.claude-plugin/plugin.json`, NOT `~/.claude/settings.json`.
+
+### 7.1 Hook wiring in plugin.json
+Read `.claude-plugin/plugin.json` and verify all 4 hooks:
+- `SessionStart` → `lcm restore`
+- `UserPromptSubmit` → `lcm user-prompt`
+- `PreCompact` → `lcm compact`
+- `SessionEnd` → `lcm session-end`
+
+Pass if: all 4 present with correct commands.
+
+### 7.2 SessionStart live test
+```bash
+echo '{}' | node dist/bin/lcm.js restore
+```
+Pass if: returns `<memory-orientation>` block.
+
+### 7.3 UserPromptSubmit live test
+```bash
+node -e 'console.log(JSON.stringify({prompt:"what changes were made to the summarizer",cwd:process.cwd()}))' | node dist/bin/lcm.js user-prompt
+```
+Pass if: returns `<memory-context>` block with hints.
+**Known issue (Bug 1):** Currently only searches promoted store. If empty, record as ⚠️ KNOWN.
+
+### 7.4 UserPromptSubmit daemon endpoint
+```bash
+node .claude-plugin/skills/lcm-dogfood/scripts/prompt-search-test.js "summarizer"
+```
+Pass if: returns hints (may be empty — see Bug 1).
+
+### 7.5 Hook timeout
+```bash
+time node -e 'console.log(JSON.stringify({prompt:"test",cwd:process.cwd()}))' | node dist/bin/lcm.js user-prompt
+```
+Pass if: completes in < 5 seconds.
+
+### 7.6 SessionEnd wiring (read-only)
+Cannot trigger without ending session. Verify wiring in plugin.json (covered by 7.1).
+
+## Phase 8: MCP Tools — 8 checks
+
+Call lcm MCP tools directly from the session.
+
+### 8.1 lcm_doctor via MCP
+Pass if: returns diagnostic results.
+
+### 8.2 lcm_stats via MCP (verbose: true)
+Pass if: returns stats with counts.
+
+### 8.3 lcm_search (query: "summarizer")
+Pass if: episodic results > 0.
+
+### 8.4 lcm_grep (query: "compact", scope: "all")
+Pass if: returns matching entries.
+
+### 8.5 lcm_store (text: "dogfood test memory — <date>", tags: ["dogfood","test"])
+Pass if: returns stored UUID.
+
+### 8.6 lcm_search retrieval ("dogfood test memory")
+Pass if: memory from 8.5 appears in results.
+
+### 8.7 lcm_expand (nodeId from prior results)
+Pass if: returns expanded content. ⚠️ SKIP if no summary IDs.
+
+### 8.8 lcm_describe (same nodeId)
+Pass if: returns metadata. ⚠️ SKIP if no node.
+
+## Phase 9: Resilience — 3 checks
+
+### 9.1 Kill daemon
+```bash
+pkill -f "lcm.*daemon" || true
+sleep 1
+node dist/bin/lcm.js status
+```
+Pass if: reports daemon down (no crash/hang).
+
+### 9.2 Auto-recovery
+```bash
+node dist/bin/lcm.js daemon start --detach
+sleep 2
+node dist/bin/lcm.js status
+```
+Pass if: daemon back up.
+
+### 9.3 Graceful degradation
+Kill daemon, then:
+```bash
+timeout 10 sh -c 'node -e "console.log(JSON.stringify({prompt:\"test\",cwd:process.cwd()}))" | node dist/bin/lcm.js user-prompt'
+```
+Pass if: returns within 10s, no crash. Restart daemon after.
+
+## Phase 10: Debug Diagnostics — 4 checks
+
+### 10.1 Daemon logs
+```bash
+tail -50 ~/.lossless-claude/daemon.log
+```
+Pass if: no ERROR entries.
+
+### 10.2 PWD matches cwd
+```bash
+echo "PWD=$PWD" && echo "cwd=$(pwd)"
+```
+Pass if: identical.
+
+### 10.3 Project DB exists
+```bash
+ls -la ~/.lossless-claude/projects/*/lcm.db 2>/dev/null
+```
+Pass if: at least one .db file exists.
+
+### 10.4 DB integrity
+```bash
+node .claude-plugin/skills/lcm-dogfood/scripts/db-integrity.js
+```
+Pass if: all DBs report "ok".

package/.claude-plugin/skills/lcm-dogfood/references/known-issues.md

@@ -0,0 +1,11 @@
+# lcm Dogfood — Known Issues
+
+Track bugs at: `.xgh/plans/2026-03-22-dogfood-findings.md`
+
+| Bug | Summary | Affects | Status |
+|-----|---------|---------|--------|
+| 1 | prompt-search only queries promoted store, not summaries or messages | Check 7.3, 7.4 | Open |
+| 2 | `env: node: No such file or directory` in daemon.log — PATH not set in spawned processes | Check 10.1 | Open |
+| 3 | No request logging in daemon — zero visibility into runtime requests | Check 10.1 | Open |
+| 4 | Config file has stale restoration fields (semanticTopK vs promptSearchMaxResults) | N/A (defaults work) | Low |
+| 5 | Skill checked settings.json instead of plugin.json for hooks | Check 7.1 | Fixed |

package/.claude-plugin/skills/lcm-dogfood/scripts/db-integrity.js

@@ -0,0 +1,40 @@
+#!/usr/bin/env node
+// Check integrity of all lcm project databases
+import { DatabaseSync } from "node:sqlite";
+import fs from "node:fs";
+import path from "node:path";
+import os from "node:os";
+
+const projectsDir = path.join(os.homedir(), ".lossless-claude", "projects");
+if (!fs.existsSync(projectsDir)) {
+  console.log("No projects directory found");
+  process.exit(0);
+}
+
+const dirs = fs.readdirSync(projectsDir).filter((d) => {
+  const dbPath = path.join(projectsDir, d, "lcm.db");
+  return fs.existsSync(dbPath);
+});
+
+if (dirs.length === 0) {
+  console.log("No project databases found");
+  process.exit(0);
+}
+
+let allOk = true;
+for (const d of dirs) {
+  const dbPath = path.join(projectsDir, d, "lcm.db");
+  try {
+    const db = new DatabaseSync(dbPath);
+    const result = db.prepare("PRAGMA integrity_check").get();
+    const status = result.integrity_check === "ok" ? "ok" : "FAIL";
+    if (status !== "ok") allOk = false;
+    console.log(`${d.slice(0, 16)}...  ${status}`);
+    db.close();
+  } catch (e) {
+    console.log(`${d.slice(0, 16)}...  ERROR: ${e.message}`);
+    allOk = false;
+  }
+}
+
+process.exit(allOk ? 0 : 1);

package/.claude-plugin/skills/lcm-dogfood/scripts/prompt-search-test.js

@@ -0,0 +1,35 @@
+#!/usr/bin/env node
+// Test the daemon's /prompt-search endpoint directly
+import http from "node:http";
+const query = process.argv[2] || "summarizer";
+const cwd = process.argv[3] || process.cwd();
+const data = JSON.stringify({ query, cwd });
+
+const req = http.request(
+  {
+    hostname: "127.0.0.1",
+    port: 3737,
+    path: "/prompt-search",
+    method: "POST",
+    headers: { "Content-Type": "application/json", "Content-Length": Buffer.byteLength(data) },
+  },
+  (res) => {
+    let body = "";
+    res.on("data", (c) => (body += c));
+    res.on("end", () => {
+      console.log(`Status: ${res.statusCode}`);
+      try {
+        const parsed = JSON.parse(body);
+        console.log(`Hints: ${parsed.hints?.length ?? 0}`);
+        if (parsed.hints?.length > 0) {
+          parsed.hints.forEach((h, i) => console.log(`  ${i + 1}. ${h.slice(0, 120)}`));
+        }
+      } catch {
+        console.log(`Raw: ${body}`);
+      }
+    });
+  }
+);
+req.on("error", (e) => console.log(`Error: ${e.message}`));
+req.write(data);
+req.end();

package/.claude-plugin/skills/lcm-e2e/SKILL.md

@@ -0,0 +1,61 @@
+---
+name: lcm-e2e
+description: Run the E2E validation checklist against a real lcm installation. Tests daemon, hooks, import, compact, promote, retrieval, and resilience. Pass a flow name for a subset (import, compact, promote, curate, retrieval, hooks, doctor, cleanup), or omit for the full suite.
+allowed-tools: Bash, Read
+user-invocable: true
+---
+
+## Current state
+
+### lcm binary
+!`lcm --version 2>/dev/null || echo "ERROR: lcm not in PATH"`
+
+### Daemon status
+!`lcm status 2>/dev/null || echo "Daemon not running"`
+
+## Task
+
+Run the E2E validation checklist from `.claude-plugin/skills/lcm-e2e/checklist.md`.
+
+**Arguments:** `$ARGUMENTS`
+
+- If `$ARGUMENTS` is empty → run all flows in order
+- If `$ARGUMENTS` is `import` → run Flows 1, 2, 3, 4
+- If `$ARGUMENTS` is `compact` → run Flows 1, 2, 5
+- If `$ARGUMENTS` is `promote` → run Flows 1, 2, 5, 6
+- If `$ARGUMENTS` is `curate` → run Flows 1, 2, 5, 6, 7
+- If `$ARGUMENTS` is `retrieval` → run Flows 1, 2, 5, 6, 8
+- If `$ARGUMENTS` is `hooks` → run Flows 1, 2, 9, 10, 14, 15, 16, 18
+- If `$ARGUMENTS` is `doctor` → run Flows 1, 11, 12, 19
+- If `$ARGUMENTS` is `cleanup` → run Flow 13 only
+
+## Steps
+
+1. Read the checklist: `.claude-plugin/skills/lcm-e2e/checklist.md`
+2. Determine which flows to run based on arguments
+3. For each step in each flow:
+   a. Print the step description
+   b. Run the command
+   c. Evaluate against expected output
+   d. Mark ✓ (pass) or ✗ (fail) with a one-line explanation
+4. If a flow fails a critical step, note it and continue (don't abort)
+5. Always run Flow 13 (cleanup) at the end of the full suite
+6. Print the summary table with actual ✓/✗ results
+
+## CRITICAL: Data Isolation
+
+**Live mode NEVER touches user data.**
+
+All operations use an isolated temp directory as `cwd`:
+- Create: `mktemp -d /tmp/lcm-e2e-test-XXXXXX`
+- This temp cwd creates a separate project database under `~/.lossless-claude/projects/<hash>/`
+- The user's real project data is never accessed
+- Cleanup removes both the temp dir AND the project under `~/.lossless-claude/projects/`
+
+## Important notes
+
+- This modifies ONLY temp/test directories — never the user's real project
+- Auto-heal flow (16) is READ-ONLY in live mode: verify hooks exist, don't break them
+- If lcm is not in PATH, show an error and stop
+- If daemon is not running, try `lcm daemon start --detach` first
+- Do not abort on first failure — the goal is a complete picture