memory-consolidate
Skill2 starsChanged 5 months agoMITView on GitHub
Reflective pass over the auto-memory tier. Surveys daily logs, decision-log entries, and recent agent invocations from the last N days, identifies durable facts that should persist as auto-memories, and proposes additions, updates, or deletions to MEMORY.md. The mechanism that keeps the auto/ tier alive.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- stoa-platform/stoatheir own version, first seen Nov 1, 2026
- YYfather/DSH-memorytheir own version, first seen Oct 15, 2026
- opensite-ai/opensite-skillstheir own version, first seen Oct 13, 2026
- mikeyobrien/rhotheir own version, first seen Oct 8, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- mikeyobrien/rho★ 372same name
- opensite-ai/opensite-skills★ 15same name
- stoa-platform/stoa★ 9same name
- YYfather/DSH-memory★ 0same name
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 repository24 other files it gives its agents.
- AGENTS.mdAGENTS.md
- control-plane/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- control-plane/CLAUDE.mdCLAUDE.md
- agentic-os-reference-syncSkill
- branch-cleanupSkill
- capture-triageSkill
- darwin-housekeepingSkill
- darwin-path-modeSkill
- decision-log-entrySkill
- eodSkill
- expert-interview-guideSkill
- feedback-intakeSkill
- harness-onboardingSkill
- ingest-contentSkill
- meeting-to-work-itemsSkill
- morning-briefSkill
- os-bootstrap-extendSkill
- os-bootstrapSkill
- pr-reviewSkill
- scheduleSkill
- spec-cross-checkSkill
- ux-critiqueSkill
- weekly-retroSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 23
After a milestone (project closure, sprint end) when patterns crystallized.
- Distinctive: line 24
When the operator notices "I've been correcting Claude on the same thing repeatedly" — that signal belongs in auto-memory.
- Very specific: line 25
When auto/ has been static for > 30 days and the system has been active.
- Distinctive: line 97
Write or update files per confirmation
- Very specific: line 98
Update MEMORY.md index — add a line for each new file, remove lines for deleted files
“Common” is measured: kept by at least 6 in 10 of the 3 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
Skill: memory-consolidate
Preconditions
control-plane/memory/auto/MEMORY.mdexists (created by os-bootstrap).- The system has at least 7 days of activity (daily logs, decision-log entries, agent calls). Without accumulated signal, consolidation is speculation.
- The operator is available for confirmation — this skill proposes; it does not silently rewrite memory.
When to use
- Weekly evolution ritual (alongside Darwin deep mode).
- After a milestone (project closure, sprint end) when patterns crystallized.
- When the operator notices "I've been correcting Claude on the same thing repeatedly" — that signal belongs in auto-memory.
- When
auto/has been static for > 30 days and the system has been active.
NOT for:
- Adding a specific fact the operator wants saved right now → direct edit + decision-log if structural.
- Cleaning up stale memories without new signal → drift correction, different shape.
- Operating-state context (active threads, current sprint) → that's
state.md, not auto-memory.
Input contract
window_days: int # default 7 — how far back to scan
include_dailies: bool # default true
include_decision_log: bool # default true
include_agent_calls: bool # default true — needs PostToolUse hook populated
include_fires: bool # default false — too noisy unless a specific pattern is suspected
domain_focus: str | None # default null = all domainsOutput contract
proposals: list[Proposal] # ≤ 5 additions / updates / deletions
# each: {action, target, body, why, confidence}
unchanged_baseline: list[str] # 3-5 things that did NOT need change — sanity check
review_summary: str # one paragraph TL;DRSequence
Step 1 — Read the inputs
control-plane/memory/daily/*.md— lastwindow_daysfilescontrol-plane/memory/decisions/decision-log.md— entries with date in windowcontrol-plane/memory/observability/agent-calls.jsonl— filter byts >= window_startcontrol-plane/memory/auto/*.md— current auto-memory inventory
Step 2 — Detect patterns
For each candidate signal, check whether it should become or update an auto-memory:
| Signal | Memory type to consider |
| Recurring correction in dailies ("Claude opens with preamble") | feedback |
| Stable fact about the operator ("works at X, focus is Y") | user |
| Active project with multi-week duration + clear deadline | project |
| Repeated pointer to an external system (Notion DB, Slack channel) | reference |
Step 3 — Apply the duplicate test
For each proposed addition:
- Is there already an auto-memory covering this? → update, don't duplicate.
- Is the signal observed only once in the window? → defer, not enough evidence.
- Is the signal structural ("I prefer X") or ephemeral ("today X was tricky")? → only structural earns persistence.
Step 4 — Apply the staleness test
For each existing auto-memory:
- Last touched > 60 days ago? → flag for re-confirmation.
- Contradicted by recent signal? → flag for update.
- Project-type memory with past deadline? → flag for removal or archival.
Step 5 — Generate proposals (≤ 5)
Each proposal:
- Action:
add/update/remove - Target: file path (existing or proposed)
- Body: the actual memory content (frontmatter + body, ready to paste)
- Why: what signal in the window triggered this
- Confidence:
high(multiple signals) /medium(one strong signal) /low(defer if not high)
Step 6 — Present + confirm
Show all proposals to the operator. Wait for confirmation per proposal:
- ✅ apply
- ✏️ apply with edit
- ❌ reject (don't repropose unless new signal)
Step 7 — Apply
- Write or update files per confirmation
- Update
MEMORY.mdindex — add a line for each new file, remove lines for deleted files - Log a decision-log entry if any proposal is structural (
type: methodologyorarchitecture)
Anti-patterns
- Acting without operator confirmation — auto-memory is the operator's voice persisted. Silent rewrites erode trust.
- Proposing > 5 changes — review fatigue degrades signal. If 5 doesn't cover it, do another pass next week.
- Promoting one-time signals — auto-memory is for stable facts. Run-of-the-mill observations belong in dailies.
- Forgetting the staleness pass — additions without subtractions bloat the tier into shelfware.
Relationship to Darwin
Darwin deep mode reads the auto-memory tier when synthesizing OS health proposals. A healthy memory-consolidate cadence (weekly or biweekly) keeps Darwin's input fresh. If memory-consolidate hasn't run in > 30 days, Darwin will flag it as a coverage gap.
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.

