diff --git a/docs/superpowers/specs/2026-05-12-memory-layer-round3.md b/docs/superpowers/specs/2026-05-12-memory-layer-round3.md
new file mode 100644
index 00000000..652a2c61
--- /dev/null
+++ b/docs/superpowers/specs/2026-05-12-memory-layer-round3.md
@@ -0,0 +1,263 @@
+# PaperForge v1.5.7 — Memory Layer Round 3
+
+> **Branch:** `feature/memory` | **Date:** 2026-05-12
+
+## Feature 1: Logging Skill — Strict Markdown Template
+
+**Problem:** Agent-written reading-log.md may not parse reliably if format varies.
+
+**Solution:** SKILL.md instructs agent to use strict template format.
+
+### File: `paperforge/skills/logging/SKILL.md`
+
+Update the reading-log route section to require this exact format:
+
+```markdown
+## ABCDEFGH — Author Last Name et al. Year
+**Title:** Full Paper Title
+
+### Section Name — Page NN or line NN-NN
+**Info:** "verbatim excerpt from paper"
+**Use:** how this supports current writing task
+**Note:** optional cross-validation note
+
+### Another Section — Page NN
+**Info:** "..."
+**Use:** ...
+**Note:** (optional)
+```
+
+### Parsing Rules (for --validate and --import):
+
+```
+paper format:    ^## [A-Z0-9]{8} — .+$         (key is 8 uppercase alphanumeric)
+title format:    ^\*\*Title:\*\* .+$
+section format:  ^### .+$
+info format:     ^\*\*Info:\*\* .+$
+use format:      ^\*\*Use:\*\* .+$
+note format:     ^\*\*Note:\*\* .+$             (optional)
+```
+
+Constraint: `info` and `use` are mandatory for every section entry. `note` is optional.
+
+### CLI Changes
+
+Update `reading-log` parser in `cli.py` to add `--validate` and `--import` subcommands under a shared parser.
+
+## Feature 2: reading-log --validate
+
+**File:** `paperforge/commands/reading_log.py`
+
+```
+paperforge reading-log --validate path/to/reading-log.md
+```
+
+Function: `validate_reading_log(filepath: Path) -> dict`
+
+Returns:
+```json
+{
+  "ok": true,
+  "file": "Project/综述写作/reading-log.md",
+  "errors": [],
+  "papers_found": 3,
+  "entries_found": 7
+}
+```
+
+On failure:
+```json
+{
+  "ok": false,
+  "errors": [
+    {"line": 15, "field": "info", "message": "missing **Info:** after section header"},
+    {"line": 23, "field": "key", "message": "paper key must match ^[A-Z0-9]{8}$"}
+  ]
+}
+```
+
+Validation algorithm:
+1. Parse into papers by `## KEY — Author` headers
+2. For each paper: verify `**Title:**` follows
+3. For each section `### ...`: verify `**Info:**` and `**Use:**` follow
+4. Report all errors at once, not stop-at-first
+
+## Feature 3: reading-log --import
+
+**File:** `paperforge/commands/reading_log.py` + `paperforge/memory/events.py`
+
+```
+paperforge reading-log --import path/to/reading-log.md
+```
+
+Function: `import_reading_log(vault: Path, filepath: Path) -> dict`
+
+Returns:
+```json
+{
+  "ok": true,
+  "papers_imported": 3,
+  "entries_imported": 7,
+  "skipped": 0
+}
+```
+
+Algorithm:
+1. Call `validate_reading_log(filepath)` — abort if errors
+2. Parse valid file into paper-level entries
+3. For each entry, call `write_reading_note(vault, paper_id, section, excerpt, usage, note)`
+4. Each write INSERTs a new row — safe for accumulative use
+
+### Add to `paperforge/memory/events.py`:
+
+```python
+def import_reading_log(vault: Path, filepath: Path) -> dict:
+    """Parse a reading-log.md and bulk-write to paper_events."""
+    # Parse, validate, write
+    ...
+```
+
+## Feature 4: reading-log --lookup KEY
+
+**File:** `paperforge/commands/reading_log.py`
+
+```
+paperforge reading-log --lookup KEY [--json]
+```
+
+Function: `lookup_paper_events(vault: Path, key: str) -> dict`
+
+Returns all accumulated paper_events for a paper, ordered by created_at DESC:
+```json
+{
+  "ok": true,
+  "zotero_key": "ABCDEFGH",
+  "title": "...",
+  "entries": [
+    {
+      "created_at": "2026-05-12 14:30",
+      "section": "Results P6",
+      "excerpt": "...",
+      "usage": "F 段参数数据",
+      "note": "与 Lippiello 对比"
+    }
+  ],
+  "count": 5,
+  "projects": ["综述写作", "数据分析"]
+}
+```
+
+## Feature 5: /methodology Skill
+
+**File:** `paperforge/skills/methodology/SKILL.md`
+
+Pure-prompt skill, no Python code. Same universal pattern as grill-me.
+
+```yaml
+---
+name: methodology
+description: >
+  Project methodology extraction. Triggered by:
+  methodology, /methodology, 提取方法论, 存档写作规律,
+  总结本项目方法, 提取可复用规则.
+source: paperforge
+---
+```
+
+### Agent workflow:
+
+1. Ask user which project to extract from (or detect from context)
+2. Read `Project/<project>/working-log.md`
+3. Identify extractable patterns:
+   - Sections marked as "方法论" or "复用"
+   - Wrong turns + corrections (弯路 + 修正)
+   - Final logic flows (最终逻辑: XX 段)
+   - Review feedback patterns (审阅修正)
+   - Cross-study audit methodology
+4. Classify into categories:
+   - `review-writing.md` — 综述写作相关
+   - `data-analysis.md` — 数据分析相关
+   - `general-methods.md` — 通用方法
+5. Present draft to user for confirmation
+6. Write to `<system_dir>/PaperForge/methodologies/<category>.md`
+
+### Methodology file format:
+
+```markdown
+---
+project: 综述写作
+extracted: 2026-05-12
+category: review-writing
+---
+
+# [Method Name]
+
+## Source
+From working-log.md Section [X.Y]
+
+## Pattern
+[Extracted reusable methodology]
+
+## Example
+[Concrete example from the project]
+```
+
+## Feature 6: Dashboard → SQLite Migration
+
+**File:** `paperforge/commands/dashboard.py`
+
+Current `_gather_dashboard_data()` does file scanning with regex. Migrate to:
+
+```python
+def _gather_dashboard_data(vault: Path) -> dict:
+    # Try DB first
+    data = _dashboard_from_db(vault)
+    if data is not None:
+        data["permissions"] = _check_permissions(vault)
+        return data
+    # Fallback to file scanning
+    return _dashboard_from_files(vault)
+```
+
+`_dashboard_from_db()` should read from paperforge.db:
+- Paper count: `SELECT COUNT(*) FROM papers`
+- Domain counts: `SELECT domain, COUNT(*) FROM papers GROUP BY domain`
+- PDF/OCR health: from papers table `ocr_status`, `has_pdf` columns
+- Remove the `_source` key (was added in earlier iteration but caused contract issues)
+
+**Keep the permissions check** (`_check_permissions`) separate and lightweight.
+
+## Feature 7: Bootstrap Update
+
+**File:** `paperforge/skills/literature-qa/scripts/pf_bootstrap.py`
+
+If present in repo (not already done), ensure `memory_layer` field is in bootstrap output. Already implemented in earlier harness work — verify status.
+
+## Refactoring: Memory Layer No Longer Optional
+
+**File:** `paperforge/plugin/main.js` ✓ DONE
+
+Removed the Easy Memory Layer toggle. Status display always shown. Memory layer is always on — SQLite is lightweight enough to not need a toggle.
+
+## Implementation Order
+
+1. Logging SKILL.md format update
+2. reading-log --validate CLI
+3. reading-log --import CLI + events.py
+4. reading-log --lookup CLI
+5. /methodology SKILL.md
+6. Dashboard SQLite migration
+7. Integration test + deploy
+
+## Cross-File Impact
+
+| File | Action | Features |
+|------|--------|----------|
+| `paperforge/skills/logging/SKILL.md` | Modify | Feature 1 |
+| `paperforge/commands/reading_log.py` | Modify | Features 2, 3, 4 |
+| `paperforge/memory/events.py` | Modify | Feature 3 |
+| `paperforge/cli.py` | Modify | Features 2, 3, 4 |
+| `paperforge/skills/methodology/SKILL.md` | Create | Feature 5 |
+| `paperforge/skills/methodology/scripts/pf_bootstrap.py` | Copy | Feature 5 (same bootstrap) |
+| `paperforge/commands/dashboard.py` | Modify | Feature 6 |
+| `paperforge/plugin/main.js` | ✓ DONE | Refactoring |