dream
Skill0 starsChanged 5 months agoMITView on GitHub
Run the Dream Mode auto-fix pass on demand (catch-up). Pulls the vault, creates today's daily note, auto-applies deterministic hygiene fixes, then resolves the remaining findings with judgment (dangling links, frontmatter, tags, dead files), writes the Morning Brief's Yesterday + Revisit parts, flips the manifest, self-checks, commits, and pushes. Use when the session-start status says last night's dream didn't run.
- Commits and pushes
42 other projects use a version of this.
Adopted by
Most recently taken up by
- borkweb/skillstheir own version, first seen Nov 15, 2026
- srinator22/agent-kerneltheir own version, first seen Nov 15, 2026
- fjwood69/moritheir own version, first seen Nov 10, 2026
- srinator22/New-Projecttheir own version, first seen Nov 8, 2026
- Melvynx/ai-maxxing-skillstheir own version, first seen Nov 8, 2026
- alexandrosandre/agentic-engineeringtheir own version, first seen Nov 8, 2026
Used with
Projects with it also use ingest, resume and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
JaimeGBASkillingest
Digest an external source (article, paper, transcript, URL, book chapter) and integrate its content into the knowledge base
Also in 4 projects that use this
genomewalkerSkillresume
Resume a thread by loading its ~800-token context capsule
Also in 3 projects that use this
MicrosoftSkillinit
Also in 4 projects that use this
Hugging FaceSkillsimplify
Refactor code after tests pass. The "Refactor" phase of Red-Green-Refactor.
Also in 3 projects that use this
SentrySkilldeslop
Strip AI-generated noise from the diff before pushing — extra comments a human wouldn't write, defensive try/catch in trusted code paths, casts to any, inline imports in Python, and other style inconsistencies with the surrounding file. Use this immediately before commit so the diff stays clean.
Also in 3 projects that use this
AnthropicSkillstart
Set up your bio-research environment and explore available tools. Use when first getting oriented with the plugin, checking which literature, drug-discovery, or visualization MCP servers are connected, or surveying available analysis skills before starting a new project.
Also in 3 projects that use this
Other versions42 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 42
- Noizefield/audio-plugin-coder★ 334same name
- grandamenium/dream-skill★ 136same name
- sneg55/agent-starter★ 76same name
- oprogramadorreal/optimus-claude★ 75same name
- tranhieutt/software_development_department★ 71same name
- Melvynx/ai-maxxing-skills★ 46same name
- maxvaega/awesome-skills★ 40same name
- seldonframe/seldonframe★ 40same name
- hyperb1iss/hyperskills★ 36same name
- fjwood69/mori★ 32same name
32 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- audit-extremeSkill
- auditSkill
- capture-noteSkill
- capture-taskSkill
- councilSkill
- daily-catchupSkill
- dailySkill
- project-scaffoldSkill
- projectSkill
- setupSkill
- weeklySkill
- workspaceSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
18 rules. 0 are common, 2 are distinctive, and 16 are very specific to this project.
0
Common
Most of the 41 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
16
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 37
If --dry-run, report what it printed and stop — do not run Layer B.
- Very specific: line 53
Never edit _meta/profile.md or _meta/memory.md.
- Very specific: line 55
Never hard-delete a file — archive to 40-archive/ instead.
- Very specific: line 61
Flow triage is proposals only: Never move an inbox item out of 00-inbox/, and never archive a note.
- Very specific: line 67
Connections are proposals only: Never add a [[wikilink]] to any note, never create a MOC / map note, and never merge, rewrite, or delete a …
- Very specific: line 78
Flip
hygiene-bin the manifest: After draining the hygiene queue (step 3), edit the > 🌙 manifest v1 | line in today's ## 🌙 Overnight … - Very specific: line 93
Leave Stuck: and Deadlines: (Layer-A facts) untouched.
- Very specific: line 95
Flip
brief-bin the manifest: After writing the brief parts, edit the > 🌙 manifest v1 | line: change brief-b:pending → brief-b:ok. - Very specific: line 103
active→shipped: when the project says it is delivered/closed/"effectively complete" (e.g.
- Distinctive: line 105
active→parked: when an unpark trigger is plainly dead or the line says it is parked.
- Very specific: line 108
If nothing qualifies, write - Proposed: — none.
- Very specific: line 121
Propose only — never move the file out of
00-inbox/: Leave the - Stale: lines (Layer A's computed orphan facts) untouched. - Distinctive: line 141
If an entry point already exists or it's not worth it, delete the line.
- Very specific: line 164
If a stage is not done, set it back to pending so the morning /daily can finish it.
- Very specific: line 167
Verify + commit: Run python .claude/hooks/validate_note.py --scan.
- Very specific: line 170
If there's no remote configured, dream.py already degrades to commit-only — just commit.
- Very specific: line 178
All deterministic file/git work happens in dream.py.
- Very specific: line 180
If dream.py exits non-zero, report and stop — do not retry blindly.
“Common” is measured: kept by at least 6 in 10 of the 41 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Dream Skill (autonomous auto-fix)
What this does
Runs the Dream Mode auto-fix pass. An optional nightly scheduler can run /dream
autonomously while you sleep (see docs/dream-autonomous-setup.md); otherwise run it
manually as a catch-up any time — this skill is the same entrypoint either way, and
nothing about the scheduler is required. Two layers:
- Layer A —
dream.py(deterministic): pull + self-re-exec, create today's note (carry-forward + week-focus), apply the no-judgment fixes (markdown→wikilink, missingdate:), write the## 🌙 Overnightsection (including the manifest header and the### Morning Briefblock withYesterday:/Revisit:placeholders and Layer-A facts forStuck:/Deadlines:), commit + push. - Layer B — you (judgment): drain the
### Needs your OKworklist with judgment, honoring the floors; write the Morning Brief'sYesterday:recap andRevisit:pick; propose orient.md section moves into the### orient.mdProposed:line (never editingorient.mditself); propose a destination for each### Flow triageinbox item (never moving the file); curate the### Connectionsfeed into worded Link / MOC / Merge proposals (never adding a link, creating a MOC, or merging a note); fliphygiene-b,brief-b,orient-b,flow-b, andconnect-bin the manifest; self-check; commit + push.
Arguments
/dream— run the pass for today./dream --dry-run— print what Layer A would do, change nothing (skip Layer B)./dream --force— re-run even if today already has a## 🌙 Overnightsection.
Steps
- Run Layer A via Bash:
python .claude/hooks/dream.py(append--dry-run/--forceverbatim if passed). If it exits non-zero (pull failed), report and stop. If--dry-run, report what it printed and stop — do not run Layer B. - Read the
### Needs your OKworklist from today's note (50-daily/YYYY/MM/YYYY-MM-DD.md). If it is empty or— none, skip to step 4. Resolve each finding with judgment, then move it from
### Needs your OKto### Auto-appliedas a one-line done-fact (e.g.- created stub: <path>). Policy:Finding Resolution Missing type/status/tagsor no frontmatterInfer from path + content ( 10-projects/…→type: project,50-daily/…→type: daily, etc.); add minimal valid frontmatter using approved tags from_meta/tags.md.Dangling [[wikilink]]Judge: (a) path-like planned sub-page under a real folder → create a proper stub note with frontmatter + one backlink; (b) dead reference (e.g. a declined project) → remove the link or make it plain text; (c) wrong path to an existing note → repoint. Never stub a bare concept link (e.g. [[SomeConcept]]) — repoint or unlink it.Unapproved tag Remap to the nearest approved tag in _meta/tags.md; if genuinely new + useful, add it to_meta/tags.mdand note the vocabulary change loudly in the log.Whole file dead/orphaned (e.g. stale _audit-*.md)Move to 40-archive/(never delete).- Floors (never cross):
- Never edit
_meta/profile.mdor_meta/memory.md. Leave any such finding in### Needs your OKand note it is floored. - Never hard-delete a file — archive to
40-archive/instead. _meta/orient.md: Layer A (dream.py) already auto-applied the deterministic heat 🔥/🌡️ + "last touched" refresh in the Active projects block — do not redo or hand-edit those tokens. You only propose section moves (next step), written into the daily note's### orient.mdProposed:line — never editorient.mditself, and never touch its## This week's focussection (that is/weekly's).- Flow triage is proposals only. Never move an inbox item out of
00-inbox/, and never archive a note. You only propose a destination for each inbox item (next step) and the stale/orphan list is Layer A's computed facts — leave the- Stale:lines exactly as written. The actual move/archive happens only when the user OKs it (or via the future/process-inbox). Archive always means move to40-archive/, never delete. - Connections are proposals only. Never add a
[[wikilink]]to any note, never create a MOC / map note, and never merge, rewrite, or delete a note. The### Connectionsfeed is Layer A's read-only computation from the optionalgraphify-out/graph; you only curate it in place (fill each candidate's_(… pending …)_placeholder, drop noise lines) inside the daily note's dream section. Leave the>staleness banner exactly as written. Applying an OK'd connection (adding the link, creating the MOC, merging) happens later, by hand — never in this pass. Do not run/graphify— graph refresh is manual and the user's call. (Ifgraphify-out/is absent, the### Connectionsblock is- — none— nothing to curate.) - The
validate_notehook validates each edit live; if it rejects one, fix and retry.
- Never edit
- Flip
hygiene-bin the manifest. After draining the hygiene queue (step 3), edit the> 🌙 manifest v1 |line in today's## 🌙 Overnightsection: changehygiene-b:pending→hygiene-b:ok. Edit only that token; leave the rest of the line intact. If the queue had floored items you could not resolve, leave the tokenpending— never markoka stage you did not actually complete. - Write the Morning Brief (Layer-B parts). In today's
### Morning Briefblock, replace the two_(pending …)_placeholders:- Yesterday: 2–3 lines summarizing the most recent prior daily note — pull from
its
## Wins,## Notes / log, and checked- [x]tasks. Compression, not analysis. If there's nothing to pull, writeYesterday: — quiet day. - Revisit: pick ONE older note worth a second look (an orphan, a parked project,
or a note untouched for weeks that connects to current focus) and ONE open
question/prompt seeded from recent themes. One line:
Revisit: [[note]] — why · Prompt: <question>. LeaveStuck:andDeadlines:(Layer-A facts) untouched.
- Yesterday: 2–3 lines summarizing the most recent prior daily note — pull from
its
- Flip
brief-bin the manifest. After writing the brief parts, edit the> 🌙 manifest v1 |line: changebrief-b:pending→brief-b:ok. Edit only that token. If you could not fill the placeholders, leave the tokenpending. - Propose orient.md section moves (Layer-B judgment). In today's
### orient.mdblock, replace the- Proposed: _(pending …)_placeholder with section-move proposals — as plain prose /-bullets, never- [ ]checkboxes. Read the Active-projects lines and propose a move only when the line's own text makes it clear:- active→shipped when the project says it is delivered/closed/"effectively complete" (e.g. "Phases 3–4 done … confirm and move to shipped").
- active→parked when an unpark trigger is plainly dead or the line says it is parked.
Name the project and give a one-line reason, e.g.
Proposed: move **Inbox cleanup** active→shipped — line says "effectively complete; confirm and move to shipped".If nothing qualifies, write- Proposed: — none. Propose only — never move the section, never editorient.md. The user (or a later/daily) applies an OK'd move by hand. Then fliporient-bin the manifest: changeorient-b:pending→orient-b:ok(edit only that token). If you could not evaluate the moves, leave itpending.
- Triage the inbox (Layer-B judgment). In today's
### Flow triageblock, for each- Inbox: \<path>` → (pending …)line, **read that inbox note** and replace the(pending …)placeholder with ONE proposed destination + a one-line reason — as **plain prose, never a- [ ]checkbox**. Destinations:10-projects/<slug>/(active, time-bounded work),20-areas/<area>/(an ongoing responsibility),30-resources/(reference material the user didn't write), or40-archive/(already-acted-on, dead, or superseded — e.g. an idea already promoted to a workspace). Example:- Inbox: `00-inbox/2026-06-19-some-idea.md` → 40-archive/ — already promoted to a workspace (capture is done).**Propose only — never move the file out of00-inbox/.** Leave the- Stale:lines (Layer A's computed orphan facts) untouched. Then **flipflow-bin the manifest:** changeflow-b:pending→flow-b:ok(edit only that token). If there were no inbox items, the block has none to fill — still flipflow-b:ok. If you could not evaluate the inbox items, leave itpending`. - Curate the Connections feed (Layer-B judgment). In today's
### Connectionsblock, turn Layer A's candidate feed into a bounded, worded set of proposals — as plain-bullets, never- [ ]checkboxes. For each candidate line, either fill its_(… pending …)_placeholder with the judgment part, or delete the line if it's noise (a pair that's already effectively linked, a community that already has an entry point, two notes that shouldn't merge). Keep it tight — a handful of high-value proposals beats the raw feed.- Link
- Link: \A` ↔ `B` — … → (theme: pending …): replace the placeholder with the shared **theme** in a few words, e.g.- Link: `20-areas/<area>/log.md` ↔ `20-areas/<area>/README.md` — both track the same plan; don't reference each other.` - MOC
- MOC: community "X" (N notes) … → _(propose [[…]] map note?: pending …)_: decide whether the cluster deserves an entry point and, if so, propose a stub map note name to confirm (never create it), e.g.- MOC: cluster "<Topic>" (10 notes) is big enough to gather — propose a [[<topic>-moc]] map note (stub to confirm).If an entry point already exists or it's not worth it, delete the line. - Merge
- Merge: \C` ~ `D` — … → (survivor: pending …): if they really are duplicates, name the suggested **survivor**; otherwise delete the line, e.g.- Merge: `_audit-2026-06-03-1106.md` overlaps `_audit-followups-2026-05-28.md` — candidate duplicate; suggested survivor the newer audit.(Distinct dated artifacts that merely share a topic are **not** a merge — delete.) **Propose only — never add the wikilink, never create the MOC note, never merge/rewrite a note.** Leave the>staleness banner untouched. If the block is- — none(no graph or no candidates), there's nothing to curate. Then **flipconnect-bin the manifest:** changeconnect-b:pending→connect-b:ok(edit only that token). If the feed was empty /— none, still flipconnect-b:ok. If you could not evaluate the candidates, leave itpending`.
- Link
- Self-check before committing (verify, don't assume). Confirm each stage you
marked
okis actually done:
hygiene-b:ok→ queue items resolved (or honestly left floored in### Needs your OK)brief-b:ok→ both_(pending …)_placeholders are gone from### Morning Brieforient-b:ok→ the### orient.mdProposed:placeholder is gone (replaced with real move proposals or— none).flow-b:ok→ every### Flow triage- Inbox:line's_(pending …)_placeholder is gone (replaced with a destination + reason), or there were no inbox items.connect-b:ok→ every### Connectionscandidate line's_(… pending …)_placeholder is gone (filled with a theme/survivor/map-note decision or the line was deleted), or the block was- — none. If a stage is not done, set it back topendingso the morning/dailycan finish it. Never mark a stageokyou didn't actually do.
- Verify + commit. Run
python .claude/hooks/validate_note.py --scan. The findings should be zero except floored items. Then commit + push via Bash:git add -A && git commit -m "dream <date>: Layer-B fixes + brief" && git push. If there's no remote configured,dream.pyalready degrades to commit-only — just commit. If the commit/push fails, report it and continue — never block on git. - Report in a few lines: how many findings Layer A auto-applied, how many you resolved in Layer B, what (if anything) remains floored, whether the Morning Brief was written, and whether the final scan was clean.
Rules
- All deterministic file/git work happens in
dream.py; don't duplicate it. - Layer B uses judgment but stays inside the floors above; every change is git-reversible.
- If
dream.pyexits non-zero, report and stop — do not retry blindly. - Manifest token names must match exactly:
hygiene-b,brief-b,orient-b,flow-b,connect-b(case-sensitive). - The manifest line format is
> 🌙 manifest v1 | ...— edit only the specific tokens you completed; leave all other tokens intact. - The Connections feed depends on the optional
graphify-out/directory; if it's absent,/dreamdegrades gracefully (Connections =— none) — it never errors on a missing graph.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

