joseph/internal-joseph
1
0
Fork 0
Code Issues 21 Pull requests Projects Releases Packages Wiki Activity Actions

[OneDev #5] feat(tooling): AI diary — Claude's running log of improvements and discoveries #38

New issue
Open
opened 2026-05-24 04:46:41 +00:00 by joseph · 0 comments
joseph commented 2026-05-24 04:46:41 +00:00
Owner
Copy link

Imported from OneDev issue #5 (id 35181)
Original project: internal-joseph
Original state: Open
Original submitterId: 5
Original submitDate: 2026-04-08T14:43:57.729+00:00
Original lastActivity: {'date': '2026-04-08T14:43:57.729+00:00', 'description': 'opened', 'userId': 5}

Original fields:

  • Type: New Feature
  • Priority: Normal
  • Assignees: None

Idea

A structured, append-only diary written by Claude, for Claude (and Joseph), capturing per-session: improvements made, gotchas discovered, tool quirks learned, workflow wins, and "aha" moments. Distinct from raw transcripts and from Joseph's personal narrative — this is the distilled, second-person layer that compounds across sessions so future-Claude inherits hard-won knowledge.

What it captures (entry types)

  • Discovery: "Found that OneDev od_create_issue rejects fields with HTTP 406 on some projects"
  • Improvement: "Switched from Bash heredoc → Write tool for multi-line files; cuts 3 round-trips to 1"
  • Gotcha: "VXLAN containers lose in-guest IP after Docker bridge init — must set ip= in pct config"
  • Workflow: "Filing OneDev issues in parallel via single message with multiple tool calls is ~5x faster"
  • Tool quirk: "MCP od_invoke_api is the escape hatch when wrapped tools 404 (e.g. /pulls vs /pull-requests)"

How it differs from what already exists

  • MEMORY.md — current state (what's true now). Diary is event log (what was learned when). Memory should reference diary for "why."
  • agent-journal (HTML) — full raw session transcripts. Diary is the curated highlights reel.
  • dev-journal (dev.pctbin.com) — Joseph's first-person narrative of his growth. Diary is Claude's third-person knowledge log.
  • project_historian agent — documents project-level decisions. Diary documents meta-learnings about how to work, what to avoid, what to repeat.

Implementation thoughts

  • Single file: ~/.claude/diary/YYYY-MM.jsonl (one entry per discovery, monthly file rotation)
  • Entry shape: {date, session_id, type, title, body, tags, related_files}
  • Capture via a skill ai-diary-entry that Claude invokes when it notices something diary-worthy mid-session
  • End-of-session sweep via wrap-up skill: "any diary entries from this session worth promoting?"
  • Optional: render as static HTML alongside agent-journal at diary.pctbin.com (IP-restricted), grouped by tag
  • Searchable from future sessions — could expose as an MCP tool (diary_search "OneDev fields gotcha")

Why

Right now, hard-won knowledge lives in three places: Joseph's head, MEMORY.md (which is hitting size limits), and buried inside session transcripts no one re-reads. Important discoveries decay between sessions — Claude rediscovers the same gotchas. A structured, queryable diary lets discoveries compound the same way commit history compounds for code.

Open questions

  • Capture mechanism: Claude self-reports proactively, or end-of-session sweep, or both?
  • Live in dotfiles (chezmoi-managed, syncs across machines) or internal-joseph (personal repo)?
  • Should the diary feed back into Claude's context automatically (e.g. relevant entries surfaced based on current task) — or only on explicit search?

Next session handoff

  • Pick this up on a Research day (per internal-joseph#1) — the design fork (proactive vs sweep) is the load-bearing decision
  • Check existing wrap-up skill before designing — it may already capture some of this
Imported from OneDev issue #5 (id 35181) Original project: internal-joseph Original state: Open Original submitterId: 5 Original submitDate: 2026-04-08T14:43:57.729+00:00 Original lastActivity: {'date': '2026-04-08T14:43:57.729+00:00', 'description': 'opened', 'userId': 5} Original fields: - Type: New Feature - Priority: Normal - Assignees: None ## Idea A structured, append-only diary written **by Claude, for Claude** (and Joseph), capturing per-session: improvements made, gotchas discovered, tool quirks learned, workflow wins, and "aha" moments. Distinct from raw transcripts and from Joseph's personal narrative — this is the *distilled, second-person* layer that compounds across sessions so future-Claude inherits hard-won knowledge. ## What it captures (entry types) - **Discovery**: "Found that OneDev `od_create_issue` rejects `fields` with HTTP 406 on some projects" - **Improvement**: "Switched from Bash heredoc → Write tool for multi-line files; cuts 3 round-trips to 1" - **Gotcha**: "VXLAN containers lose in-guest IP after Docker bridge init — must set `ip=` in `pct config`" - **Workflow**: "Filing OneDev issues in parallel via single message with multiple tool calls is ~5x faster" - **Tool quirk**: "MCP `od_invoke_api` is the escape hatch when wrapped tools 404 (e.g. `/pulls` vs `/pull-requests`)" ## How it differs from what already exists - **`MEMORY.md`** — current *state* (what's true now). Diary is *event log* (what was learned when). Memory should reference diary for "why." - **`agent-journal` (HTML)** — full raw session transcripts. Diary is the curated highlights reel. - **`dev-journal` (dev.pctbin.com)** — Joseph's first-person narrative of his growth. Diary is Claude's third-person knowledge log. - **`project_historian` agent** — documents project-level decisions. Diary documents *meta-learnings* about how to work, what to avoid, what to repeat. ## Implementation thoughts - Single file: `~/.claude/diary/YYYY-MM.jsonl` (one entry per discovery, monthly file rotation) - Entry shape: `{date, session_id, type, title, body, tags, related_files}` - Capture via a skill `ai-diary-entry` that Claude invokes when it notices something diary-worthy mid-session - End-of-session sweep via `wrap-up` skill: "any diary entries from this session worth promoting?" - Optional: render as static HTML alongside agent-journal at `diary.pctbin.com` (IP-restricted), grouped by tag - Searchable from future sessions — could expose as an MCP tool (`diary_search "OneDev fields gotcha"`) ## Why Right now, hard-won knowledge lives in three places: Joseph's head, MEMORY.md (which is hitting size limits), and buried inside session transcripts no one re-reads. Important discoveries decay between sessions — Claude rediscovers the same gotchas. A structured, queryable diary lets discoveries *compound* the same way commit history compounds for code. ## Open questions - Capture mechanism: Claude self-reports proactively, or end-of-session sweep, or both? - Live in `dotfiles` (chezmoi-managed, syncs across machines) or `internal-joseph` (personal repo)? - Should the diary feed back into Claude's context automatically (e.g. relevant entries surfaced based on current task) — or only on explicit search? ## Next session handoff - Pick this up on a **Research day** (per internal-joseph#1) — the design fork (proactive vs sweep) is the load-bearing decision - Check existing `wrap-up` skill before designing — it may already capture some of this
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
No results found.
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
joseph/internal-joseph#38
No description provided.