Run this skill when the user asks to improve, audit, or maintain memory quality.
Strict Runbook (for weaker agents)
When this skill is triggered, execute in this order:
- Read current state files:
USER.mdTASKS.mdmemory/YYYY-MM-DD.md(today)- optional:
agent-memory/common-info.md
- Set retrieval scope before searching (when applicable):
- user / project / session / date range
- avoid global scans if a narrower scope is enough.
- Retrieve with efficiency gates (must follow):
- L1: today memory + task board (+
memory/keyword-frequency.mdfor keyword questions) - run a sufficiency check: if answer is already supported by evidence, stop here
- L2: scan
memory/*.mdonly when L1 is insufficient - run a sufficiency check again
- L3: fallback to
agent-memory/chat/daily -> weekly -> monthlyonly when confidence remains low
- L1: today memory + task board (+
- For every important new decision, append one structured block:
Decision / Why / Impact / Next / Verify- optional but recommended:
RefID / Scope / Confidence / Tags / Aliases
- Update
TASKS.mdwith only在做 / 就绪 / 中断.- If
中断, must fillBlocked byandResume step.
- If
- After memory write-back, update
memory/keyword-frequency.md. - Apply checkpoint policy (for
/new-resilient mode):- Event trigger (mandatory): on new preference/boundary, task state change, or key external action completion, append write-back immediately.
- Pre-refresh trigger (mandatory): before
/new//reset/ context refresh, append a lightweight checkpoint first. - Periodic fallback (optional but recommended): every 30–45 minutes, append a lightweight checkpoint with a min-interval guard (e.g., skip if last checkpoint < 25 minutes).
- If no new decisions, still write a minimal checkpoint (timestamp + TASKS snapshot).
- If the same rule/decision repeats >=3 times, compact and promote:
- long-term preference ->
USER.md - operational rule ->
agent-memory/common-info.md - keep a reference link to original daily entries.
- long-term preference ->
- Reply with verifiable evidence (file path + what changed), not vague claims.
Never skip step 3 for historical recall questions.
Core Model
Use 3 layers:
USER.md: stable user profile and long-term preferences.memory/YYYY-MM-DD.md: daily facts and decisions.TASKS.md: active executable tasks only (在做/就绪/中断).
Use TASKS_ARCHIVE/YYYY-MM.md for completed tasks.
Daily Write-Back Standard
For key events, write:
DecisionWhyImpactNextVerify- optional:
RefID,Scope,Confidence,Tags,Aliases
When the user reports memory mismatch ("You didn't remember"), immediately show the exact stored location and verify against file content.
Checkpoint Standard (frequent /new support)
Use this lightweight checkpoint block when there is no major new decision but state continuity is needed:
RefID(e.g.,AUTO-YYYYMMDD-XX)Decision(auto flush checkpoint)Why(protect continuity before reset/switch)Impact(recoverable context after reset)Next(continue normal write-back loop)Verify:- checkpoint time
TASKS.mdsnapshot counts (在做/就绪/中断)- optional top active task IDs
Goal: preserve recoverability without bloating daily notes.
Retrieval Fallback Order
- L1: today memory + task board
- L2:
memory/*.md - L3:
agent-memory/chat/daily -> weekly -> monthly(when low confidence)
Keyword-first rule (must apply)
For keyword questions, do this before final answer:
- Read
memory/keyword-frequency.mdfor candidate terms and counts. - Locate matching daily entries in
memory/*.md. - Return concrete path evidence (where the keyword/decision appears).
Always prefer verified answers over guessed answers.
Efficiency Enhancements (memU-inspired)
- Pipeline stages, not one-shot search
- Run retrieval as staged execution (L1/L2/L3 + sufficiency checks).
- Scope guard first
- Narrow
wherescope before retrieval to reduce noise and latency.
- Narrow
- Sufficiency gate
- Stop retrieval once evidence is enough; avoid expensive unnecessary scans.
- Reference-friendly write-back
- Add short
RefIDin important entries so later answers can cite quickly.
- Add short
- Reinforcement + compaction
- Repeated decisions should be promoted to stable layers to prevent daily-file bloat.
Task Board Rules
- Keep only active states in
TASKS.md:在做/就绪/中断. - For
中断, require:Blocked byResume step
- Move completed tasks to
TASKS_ARCHIVE/YYYY-MM.md.
Rule Placement
- Keep user identity and stable preferences in
USER.md. - Keep operational rules (migration/archive/retrieval) in
agent-memory/common-info.md.
Keyword Tracking
- Maintain keyword seeds manually in
memory/keyword-seeds.md. - Generate stats to
memory/keyword-frequency.mdafter each daily write-back. - Default mode is event-driven.
- For high-frequency
/newworkflows, you may enable optional periodic checkpoint mode; if enabled, keep a min-interval guard to avoid noisy duplicate entries.
References
- Workflow details:
references/workflow.md - Daily template:
references/daily-template.md - Task board template:
references/tasks-template.md - Efficiency notes:
references/efficiency-upgrades.md - Checkpoint mode:
references/checkpoint-mode.md