omp-autodream

Archived 2026-10-03. This repo is merged into STRML/autodream (formerly cc-autodream), which reads Claude Code and OMP sessions in one nightly and writes one report. The OMP session reader is adapters/omp/ there. The commit-by-commit record of what was ported, and why a few commits were not, is in docs/plans/2026-10-03-omp-adapter.md in that repo. To move an install over, follow "Moving from omp-autodream" in its README. Issues here stay for reference; file new ones in the unified repo.

Nightly session review for the OMP harness (Oh My Pi). While you sleep it reads yesterday's session transcripts and leaves you one short report: the mistakes you keep making, where you lost time, and what's worth remembering.

This is a port of cc-autodream that runs the nightly pipeline on OMP: it reads OMP session transcripts with OMP's models and writes reports to OMP's own data directory. The interactive triage tool (review.sh) is inherited from cc-autodream and still runs on the Claude Code CLI — see review.sh below.

Why

You have dozens of OMP sessions a week. The useful lessons — a wrong assumption you made twice, a flag that always trips you up, a fix worth keeping — are buried in transcripts you'll never reread. OMP starts each session fresh and rediscovers the same friction.

omp-autodream does the rereading for you. Every night it looks across all of yesterday's sessions, ranks what recurs by frequency × severity (so you see the patterns, not the one-offs), and writes a dated digest you can skim in a minute. The report is the whole deliverable — under OMP, Layer 2 performs no memory writes, so nothing else in your session corpus is touched.

What it uses

• Session source: OMP's session store, ~/.omp/agent/sessions/<project>/<ts>_<uuid>.jsonl. OMP's transcripts carry the same substance Claude's do — user/assistant message records, custom records for tool calls, model_change provenance — so the triage reads full conversation + tool activity, not just prompts.
• Layer 1 (per-session triage): deepseek/deepseek-flash — cheap, fast. Every L1 worker runs with the advisor disabled (--config l1-no-advisor.yml), so a nightly fan-out doesn't boot the expensive advisor model on every session. The same overlay turns off first-turn memory recall, which made every worker exit 0 before calling the model.
• Layer 2 (aggregation): anthropic/claude-opus-5 on the Anthropic subscription (file-based agent.db OAuth — safe unattended at 3am; no keychain dependency).
• Changelog window: diffs the changelog of all three harnesses you work across — Claude Code, Codex and OMP — over the report's date window, and the report says what each relevant release means for your setup rather than reprinting the release notes. Each source has its own cache and its own section, so one unreachable remote never silences the others. Override the watched set with AUTODREAM_CHANGELOG_SOURCES.

What you get

A report at ~/.omp/agent/dreams/YYYY-MM-DD.md. See example/2026-05-25.md for a real one (60 sessions, seven ranked patterns, five open questions). The shape:

# Dream report — 2026-05-29   (21 sessions)

## Top patterns
1. **Editing files before reading them** ×6  ·  high
   Six sessions hit "File has not been read yet". Read-before-Edit isn't sticking.
   → Pin: always Read a file in-session before the first Edit.
2. **$() in Bash triggers permission prompts** ×4  ·  medium
3. **Assuming the host TZ is Pacific** ×2  ·  medium

## Upstream Claude Code changes
- v2.x.y added `--foo`; relevant to the rush project's deploy script.

## Open questions for the user
1. Add a Read-before-Edit reminder to the project CLAUDE.md?

How you interact with it:

• It runs unattended. A launchd job fires overnight (with morning catch-up triggers in case the Mac was asleep). You do nothing.
• It opens itself. When a report lands, notify.sh writes its open questions to a text file and pops it open in your editor — so it's in front of you with morning coffee, not waiting to be discovered.
• It gets louder when a question goes unanswered. Once the report has asked the same question 3 reports in a row, a second banner tells you it is still unanswered, and the question lands in findings/<date>/question-escalations.txt. Two questions are the same when their bold titles match word for word (case and spacing aside). A night with no report doesn't break a streak; a report that drops the question ends it. Set AUTODREAM_QUESTION_ESCALATE_AT to change the 3.
• It has an interactive suggestion solver. review.sh opens a Claude session preloaded with the report and walks the open questions one at a time — restate, recommend, then approve / modify / skip / discuss — executing the ones you approve and logging each decision back into the report.
• It stays quiet when there's nothing to say. Most nights the report has no open questions, and opening a session just to be told "nothing to do" costs a session's tokens to print one line. review.sh detects that itself and prints the line instead. Same for a report you've already triaged. Anything it can't classify still opens the session — a wasted session is cheaper than a buried question. --force overrides.

Install

git clone https://github.com/STRML/omp-autodream ~/git/omp-autodream
cd ~/git/omp-autodream
./install.sh

This symlinks bin/*.sh and prompts/*.md into ~/.omp/agent/autodream/, creates ~/.omp/agent/dreams/, and on macOS installs and bootstraps the nightly launchd schedule for you (auto-detecting your username, paths, and the omp binary — no plist editing). Because the scripts are symlinks, editing the repo copy takes effect immediately.

Install links the worker overlay (l1-no-advisor.yml) into AUTODREAM_DIR. The nightly runners pass it as --config, so no headless worker boots the advisor, probes local providers, or runs first-turn memory recall.

The schedule fires run.sh at 03:15 with morning catch-up triggers (06:15/09:15/12:15) in case the Mac was asleep; the idempotency guard makes all but the first a one-second no-op. To skip scheduling and only symlink the scripts:

./install.sh --no-schedule

launchd won't wake the Mac, so to guarantee the 03:15 trigger runs at all, add a scheduled wake:

sudo pmset repeat wake MTWRFSU 03:10:00

The hand-editable template lives at launchd/com.user.omp-autodream.plist.example if you'd rather install the job yourself.

Running it by hand

~/.omp/agent/autodream/run.sh $(date -v-1d +%Y-%m-%d)       # process yesterday
AUTODREAM_FORCE=1 ~/.omp/agent/autodream/run.sh 2026-05-29  # rebuild a date
~/.omp/agent/autodream/review.sh                            # solve the latest report's questions
~/.omp/agent/autodream/review.sh 2026-05-29                 # triage a specific report
~/.omp/agent/autodream/review.sh --force 2026-05-29         # open it even if there's nothing to triage
~/.omp/agent/autodream/question-streaks.sh status           # questions asked in consecutive reports
~/.omp/agent/autodream/question-streaks.sh clear <key>      # forget one you've acted on (clear all: every one)

$OMP_BIN overrides the omp binary location (default /opt/homebrew/bin/omp). The L1 runner model is AUTODREAM_L1_MODEL (default deepseek/deepseek-flash); the L2 aggregator model is AUTODREAM_L2_MODEL (default anthropic/claude-opus-5). All knobs are documented in bin/run.sh's header.

Interactive triage (review.sh)

review.sh is inherited from cc-autodream and still shells out to the Claude Code CLI (claude, override with CLAUDE_BIN): it runs an interactive session to walk the report's open questions. The nightly pipeline (L1/L2) is fully on omp; only the interactive solver uses claude, so it needs the Claude CLI installed.

review.sh exits without opening a session when the report has no open questions or already carries a ## Triage decisions section, printing where the report is. It reads the <!-- autodream:open-questions=N --> marker that PROMPT.md makes the nightly run emit; anything ambiguous opens the session as before.

To make it open in its own cmux workspace instead of the current terminal, drop a config at ~/.omp/agent/autodream/config (see example/config.example):

AUTODREAM_TRIAGE_SURFACE=cmux    # inline (default) | cmux

Leaving notes for the next run

Notes you leave get answered in the report's Operator notes section, with evidence from that night's sessions — "how often did I actually use /graphify, and did it work?" comes back as a count and a verdict, not a guess.

From a terminal:

~/.omp/agent/autodream/autodream-note.sh "evaluate how often /graphify is used"
~/.omp/agent/autodream/autodream-note.sh --expires 2026-10-01 "check the codemaps hook overhead"

From anywhere else, including your phone, point autodream at a folder in a synced vault and drop a markdown file in its inbox:

# in ~/.omp/agent/autodream/config — quote it, the iCloud path has spaces
AUTODREAM_VAULT_DIR="$HOME/Library/Mobile Documents/iCloud~md~obsidian/Documents/vault/autodream"

The run creates the rest on its own:

path	what it holds
<vault>/inbox/*.md	drop a note here; one file per note, any filename
<vault>/processed/<date>/	where a note moves once a report has read it
<vault>/reports/<date>.md	the nightly report, copied in so you can read it in bed

Add expires: YYYY-MM-DD as YAML frontmatter to a note that should retire itself. A note is only archived after a report actually read it, so a failed run leaves your inbox alone, and so does a note you write while a run is in flight.

Ideas from your X bookmarks

Point autodream at your X bookmarks and the report gains an Ideas from bookmarks section: what you saved, crossed against what you actually worked on that day. A bookmark that connects to nothing gets left out — the section is the intersection, not a reading list.

X's official API can't do this (the bookmarks endpoint needs the $200/mo Basic tier), so this reads the same web endpoint your browser does, with your session cookies:

1. Open x.com in a browser, logged in. DevTools → Application → Cookies → https://x.com.
2. Copy the values of auth_token and ct0.
3. Put them in ~/.omp/agent/autodream/x-credentials:

X_AUTH_TOKEN=paste_auth_token_here
X_CT0=paste_ct0_here

chmod 600 ~/.omp/agent/autodream/x-credentials
~/.omp/agent/autodream/x-bookmarks.sh status   # check it took

Bookmarks are marked read once a report has covered them, so each one gives you an idea once. Without that file the feature is simply off. When the cookies expire the report says so and tells you to re-paste; the run itself never fails over it.

Running long jobs

A full run usually takes more than 10 minutes. If you're kicking it off from something that kills long jobs (a Claude Code background task, a flaky ssh session), use autodream-now.sh — it hands the run to launchd, which has no time cap and keeps going after you disconnect:

~/.omp/agent/autodream/autodream-now.sh                    # yesterday, right now
~/.omp/agent/autodream/autodream-now.sh 2026-05-29 --force # a specific date, rebuild
~/.omp/agent/autodream/autodream-now.sh 2026-05-29 --watch # follow the log until the report lands

How it works (short version)

Two layers: a cheap per-session pass (Layer 1, deepseek/deepseek-flash) extracts structured findings from each transcript, then a single smarter pass (Layer 2, anthropic/claude-opus-5) ranks them across the whole day and writes the report. Layer 1 workers invoke omp -p headless with the advisor disabled; Layer 2 invokes omp -p on the subscription model. It also diffs the upstream Claude Code changelog over the day so the report can flag releases that change how you work.

Everything lives on disk (findings JSON, the report, run logs, stats) and every step is idempotent, so you can rerun any date. Configuration knobs are documented in bin/run.sh's header.

For the internals — data flow, file map, state layout, environment overrides, the headless-worker pattern, and the "don't eat your own tail" self-pollution defenses — see codemaps/architecture.md and CLAUDE.md.

Caveats

• macOS-only as written (BSD date, launchd, osascript/open). The core pipeline is portable; the scheduling and notify bits are mac-specific. The morning open-questions file opens with your default .md app; set AUTODREAM_OPEN (e.g. subl, code -g, open -a Obsidian) to pick a specific editor.
• The nightly pipeline runs headless omp -p workers in --approval-mode yolo (Layer 1 workers write findings; the Layer 2 aggregator writes the daily report). Don't run it in a shared environment.
• Auth: both layers use subscription OAuth via agent.db and need no API key on the default models. A run does serialize one throwaway model call before the fan-out (l1_warmup in run-stats.txt, disable with AUTODREAM_L1_WARMUP=0). It was added against a suspected cold-token race that turned out not to be the cause, and it stays because it is cheap and never fatal. If you point AUTODREAM_L1_MODEL at a keyed provider, put its key in ~/.omp/agent/autodream/x-credentials (chmod 600, key=value): run.sh sources that file and nothing else. There is no keychain fallback, despite what this bullet claimed before 2026-09-11 — a key that lives only in the keychain reaches the workers unset.
• The first run clones anthropics/claude-code (small) for the changelog window; it degrades gracefully with no git/network.

License

MIT — see LICENSE.