菜单
Derives the session's authorship tally from the conversation and the plan diff at session end. Presents draft for author approval before appending to the project's authorship_log.md, the only place authorship is recorded.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
| name | log-session |
| description | Derives the session's authorship tally from the conversation and the plan diff at session end. Presents draft for author approval before appending to the project's authorship_log.md, the only place authorship is recorded. |
This skill produces an auditable record of authorship for AI-assisted thesis writing sessions. At session end it derives the session's authorship tally from the conversation and the plan diff, presents an entry for author review, and appends the approved entry to the project's authorship_log.md.
The log records the author's decisions, rejections, and domain contributions. It is not a transcript.
authorship_log.md is the only place authorship is recorded. plan.md, evidence.md, and drafted .tex files carry no authorship or approval-stage field. Write no authorship file during the session.
git diff over the project's plan.md and evidence.md files. Gives the structural counts.authorship_log.md in the thesis project root, for the cumulative summaryDetermine the project root and confirm it is a git repository. Establish the session's baseline commit: the most recent commit made before this session's first edit, found with git log over the plan files. State the baseline explicitly in the entry.
Collect the structural change with git diff <baseline> -- '*plan.md' '*evidence.md'. If the project is not a git repository, or the working tree was already dirty at session start so no clean baseline exists, say so in the entry and derive every count from conversation alone.
Count sentence points across the session's scope:
| Count | Definition |
|---|---|
| Points recorded | Sentence points written or rewritten in the scope this session |
| Adjusted by grounding | Points whose text changed as a result of the grounding pass |
| Agent-suggested, unchallenged | Points the agent proposed that entered the plan with no author edit or objection |
| Edited or added by the author | Points the author dictated, added, reworded, or altered after an agent proposal |
The last two partition the recorded total. Grounding adjustments are orthogonal and overlap both.
Take points recorded and the added or changed subtotals from the diff. Take the split between the last two rows from the conversation; the diff cannot separate them.
The tally is a session aggregate. Do not track authorship per point, do not enumerate point IDs, and do not reconstruct a per-point history.
From the conversation, identify:
Author direction — where the author introduced a technical point, claim, or structural choice; rejected an agent suggestion, with brief reason if apparent; modified an agent suggestion before accepting; supplied domain knowledge not available in the literature; or redirected emphasis, ordering, or scope.
Agent contributions — where the agent proposed structure or content accepted without significant modification, suggested Zotero references that were accepted, or performed organisational work such as sequencing, grouping, or formatting.
Iteration indicators — scopes that required multiple revision cycles before approval, and the approximate exchange count.
## Session [DATE] — [Scope Description]
**Exchanges**: ~[N] | **Skills used**: [list]
**Diff baseline**: [commit | none, stated reason]
### Scope
[1-2 sentences: what was worked on this session]
### Authorship Tally
| Count | Value |
|--------|-------|
| Points recorded | [N] |
| Adjusted by grounding | [N] |
| Agent-suggested, unchallenged | [N] |
| Edited or added by the author | [N] |
**Summary**: [1-2 sentence plain-language interpretation of the session's authorship balance]
### Author Direction
- [Concrete decisions, rejections, and domain contributions — 3-8 bullet points]
- [Each bullet specific enough to demonstrate intellectual control]
- [Include section/paragraph references where possible]
### Agent Contributions
- [What the agent provided — structural organisation, reference suggestions, prose drafting]
- [Be honest about agent-originated content that was accepted]
### Iteration & Negotiation
- [Scopes that required significant back-and-forth]
- [Key points of disagreement and how they were resolved]
### Files Modified
- [Files written or edited during the session]
Present the draft entry as a complete block. The author will approve as-is, request specific corrections (misattributed decisions, missing context, inaccurate characterisation), or add points the log missed.
Handle corrections conversationally — update the draft and re-present until approved.
Do NOT:
Once approved:
authorship_log.md in the thesis project rootThe top of authorship_log.md contains a running summary updated each session:
# Authorship Log
## Cumulative Summary
- **Sessions logged**: [N]
- **Chapters/sections covered**: [list]
- **Total exchanges**: ~[N]
- **Tool**: Codex [model/version], thesis-writer plugin v[version]
- **Process**: All content planned collaboratively via document-planner,
prose drafted via writer skill from approved plans. All citations from
author's Zotero library. Author reviewed and approved all output.
### Cumulative Authorship Tally
| Count | Total |
|--------|-------|
| Points recorded | [N] |
| Agent-suggested, unchallenged | [N] |
| Edited or added by the author | [N] |
---
[Session entries in reverse chronological order]
.tex files, figures).tex fileThe log must be accurate, not flattering. Report the session as it happened, not as the agent wishes it had.
Interactive top-down planning at thesis, chapter, section, paragraph, and sentence scope. Use to draft author-readable plan.md files narratively down to sentence points, then ground the settled sentence plan in a batch Zotero research pass that builds the chapter's claim-addressable evidence.md ledger for author haggling.
Derives the session's authorship tally from the conversation and the plan diff at session end. Presents draft for author approval before appending to the project's authorship_log.md, the only place authorship is recorded.
Conversational technical LaTeX writer. Use after an approved author-readable plan.md and matching evidence.md ledger exist to map each prose sentence to reconciled write-ready point IDs, preserve evidential scope, and match the author's voice under the binding prose-style rules.
Interactive top-down planning at thesis, chapter, section, paragraph, and sentence scope. Use to draft author-readable plan.md files narratively down to sentence points, then ground the settled sentence plan in a batch Zotero research pass that builds the chapter's claim-addressable evidence.md ledger for author haggling.
Conversational technical LaTeX writer. Use after an approved author-readable plan.md and matching evidence.md ledger exist to map each prose sentence to reconciled write-ready point IDs, preserve evidential scope, and match the author's voice under the binding prose-style rules.
Generates figures from placeholders in .tex files. Creates Python matplotlib scripts for data figures, TikZ/Python schematics for diagrams. Maintains a shared plot_defaults.py for consistent styling. Flags complex figures for the user. Runs after the writer skill.