Monthly log processing oriented toward work consolidation. Split and archive by month from Daily Note.md, extract structured signals, and generate fact-first monthly summaries. The monthly review is Lode's month-level compounding layer: in addition to summarizing outcomes, it must also identify recurring open questions, risks, habit changes, and ways of working that need adjustment next month.
Core outputs:
- Monthly archive (raw Markdown, not rewritten)
- Monthly summary (written by the agent, fact-first, no fabrication)
- Intermediate data (signals.json + skeleton.json, for secondary consumption)
- Candidate rules (suggestions only, not automatically written to AGENTS/checklists/skills)
Design principles
- Fact-first: only extract and generalize from original records, do not fabricate missing information
- Original text unchanged: archive files retain the original Markdown without rewriting
- Summary and log separated: archives and summaries are independent files
- Summarizing allowed, fabrication prohibited: multiple records may be generalized into one theme, but nonexistent "outcomes" or "conclusions" must not be generated out of thin air
- Periodic compounding: the summary must point out what should change next month, rather than merely restating what was done this month
Configuration
This skill uses the Lode unified configuration system. Resolve the knowledge base path from the following locations ({vault}), with higher priority taking precedence:
| Priority | Location | Description |
|---|---|---|
| 1 | .lode/config.yaml(project root directory) |
Project-level override |
| 2 | ~/.lode/config.yaml |
Global configuration |
| 3 | $WEEKLY_PPT_PATH environment variable |
legacy fallback |
| 4 | ~/.weekly-ppt/ |
legacy fallback default value |
Project-level configuration overrides fields with the same name in the global configuration. This skill's main outputs depend on {vault}; if the path cannot be resolved, prompt the user to configure knowledge_vault. For the complete configuration format and merge rules, see references/weekly-ppt-convention.md。
This skill's path mapping:
- Input:
{vault}/Daily Note.md - Archive output:
{vault}/Work Diary/Monthly/{YYYY-MM}.md - Intermediate data:
{vault}/raw/months/{YYYY-MM}/signals.jsonandskeleton.json - Summary output:
{vault}/Work Diary/Monthly/{YYYY-MM}.summary.md
Input parameters
Required
| Parameter | Description |
|---|---|
input_file |
Path to Daily Note.md (derived from configuration by default) |
output_dir |
Output directory path (default {vault}/Work Diary/Monthly/) |
Optional
summary_mode: project_focused # light | project_focused | engineering_review
overwrite_policy: overwrite # append | overwrite | skip_existing
month_filter: null # 只处理指定月份,如 "2026-03"
project_filters: [] # 只保留特定项目视角
evidence_mode: strict # strict | best_effort
summary_mode:lightonly extracts the main thread and keywords;project_focused(default) merges by project;engineering_reviewstrengthens the problem/solution/risk structureevidence_mode:strict(default) each summary must have a basis in the original text,best_effortallows moderate generalization
Execution steps
Step 1: Read and pre-check
- Confirm that
input_fileexists and is readable (UTF-8) - Identify the heading hierarchy structure (year → month → date)
- If the format does not conform, prompt the user to refer to
references/daily-note-format.md
Heading regex:
年标题: ^#\s*\d{4}\s*年
月标题: ^##\s*\d{4}\s*年\s*\d{1,2}\s*月
日标题: ^###\s*\d{4}\.\d{2}\.\d{2}
Step 2: Split and archive by month
Run scripts/split_daily_note.py, splitting Daily Note.md into YYYY-MM.md monthly archive files, saved to {vault}/Work Diary/Monthly/。
- Retain the original content without rewriting
- Automatically infer from the date when the month-level heading is missing
- Supports
overwrite_policyandmonth_filter
Step 3: Extract signals + build skeleton
Run scripts/prepare_monthly_data.py(merged the original extract + build scripts):
python scripts/prepare_monthly_data.py \
--input {vault}/Work\ Diary/Monthly/{YYYY-MM}.md \
--signals-output {vault}/raw/months/{YYYY-MM}/signals.json \
--skeleton-output {vault}/raw/months/{YYYY-MM}/skeleton.json \
--summary-mode project_focused \
--evidence-mode strict
The script completes all deterministic work:
- Extract project/module/category tags, task status, daily focus, risk signals, next-step signals
- Merge by project, compute distribution statistics, identify work phases
- Automatically identify real projects (those appearing on >= 2 days)
- Output signals.json and skeleton.json
Step 4: Agent writes the summary
Read the references/worklog-summary-template.md template,skeleton.json skeleton data, raw monthly archive YYYY-MM.md。
Write {YYYY-MM}.summary.mdaccording to the template, save to {vault}/Work Diary/Monthly/。
Important constraints:
evidence_mode=strict, every summary must be traceable to a corresponding entry in the source text- Do not describe "in progress" as "completed"
- Do not package multiple independent small fixes as a "major achievement"
- When a project has only scattered records, honestly state "few records"
Trade-off rules:
- Each project has at most 2-3 core themes; merge the rest into "routine optimizations include..."
- Keep code change numbers only when net additions > 300 lines or net deletions > 200 lines
- Merge config adjustments, documentation optimization, and small fixes of < 50 lines into one sentence
- Total summary length 80-120 lines
- If multiple dates record the same workflow, merge into one theme and describe it by the latest status; do not amplify repeated traces of the same raw/git source in daily reports into multiple achievements
- If the same theme contains completed items, risk items, and decision items at the same time, organize it as "what was completed + what risks/decisions remain", avoiding a success-only narrative
- Identify recurring open questions, stale threads, unresolved risks, and explain how they affect next month's plan
- Summarize the usage of Lode habit loops such as
开工/收工: which sessions have records, and which contexts may be missing - Output candidate rules: from repeated evidence, propose AGENTS rules, checklists, playbooks, or skill ideas that may be worth solidifying. A single weak signal can only be written as an observation or checklist suggestion, and cannot be upgraded into a strong rule.
Step 5: Output execution summary
Output an execution summary in JSON format to the terminal:
{
"months_created": ["2026-03"],
"summaries_created": ["2026-03.summary.md"],
"summary_mode": "project_focused",
"projects_detected": ["项目A", "项目B"],
"warnings": []
}
Project classification rules
Projects are automatically identified via - [项目名] tags in the Daily Note. Core principles:
- Use
[项目名]tags as the basis for classification; do not guess based on content - Markdown links
[文本](URL)are not project tags - See
references/project-tagging-guide.md
Boundaries and non-goals
This skill does not handle: rewriting original Daily Notes, writing report copy, generating performance statements, or judging work value.
This skill only handles: split → archive → distill → structured output → propose candidate rules.
Candidate rules are proposal-only: unless the user explicitly requests and approves writing to the target files, AGENTS.md, checklists, playbooks, or skills must not be modified automatically.
Error handling
- File does not exist: report the path, prompt to check
- Encoding error: report the location, suggest UTF-8
- Heading level not recognized: show the actual levels, prompt to refer to
references/daily-note-format.md - A certain month has no content: Normal processing, note "No records this month" in the summary
- Script execution failed: Report the complete error, do not swallow exceptions
Reference file
| File | When to read |
|---|---|
references/daily-note-format.md |
Understand the Daily Note format conventions |
references/worklog-summary-template.md |
When writing the summary (must read) |
references/project-tagging-guide.md |
Project categorization rules |
references/weekly-ppt-convention.md |
Shared storage conventions |