Vibe-Skills deepagent-memory-fold

DeepAgent-style memory folding for VCO sessions: compress long context into structured working/tool memory without using episodic-memory.

install

source · Clone the upstream repo

git clone https://github.com/foryourhealth111-pixel/Vibe-Skills

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/foryourhealth111-pixel/Vibe-Skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/bundled/skills/deepagent-memory-fold" ~/.claude/skills/foryourhealth111-pixel-vibe-skills-deepagent-memory-fold && rm -rf "$T"

manifest: bundled/skills/deepagent-memory-fold/SKILL.md

DeepAgent Memory Fold (VCO)

When to use

Use this skill when:

The task is long-horizon and context is getting large
You need to “take a breath” and restart reasoning from a compact state
You see repeated retries / route instability / losing track of decisions
You need to hand off to another agent or start a new session

Governance constraints (must follow)

VCO memory governance disables
```
episodic-memory
```
.
Use state_store (session) by default.
Only write to Serena memory when the user explicitly approves a project decision.

Runtime (Upstream vendoring)

DeepAgent upstream is vendored (optional/advanced):

C:\Users\羽裳\.codex\_external\ruc-nlpir\DeepAgent\

Runtime config + preflight (no secrets stored/printed):

C:\Users\羽裳\.codex\skills\vibe\config\ruc-nlpir-runtime.json

pwsh C:\Users\羽裳\.codex\skills\vibe\scripts\ruc-nlpir\preflight.ps1

Output contract (structured fold)

Produce a “folded memory” object with these sections:

Working memory
- Current goal
- Current sub-goal
- Current blockers
- Next 3 actions
Tool memory
- Tools/skills used
- What worked / what failed
- Availability notes (keys required, deps missing)
Evidence memory
- Top 5 evidence anchors (file:line or URLs)
Decision log
- Only decisions actually made (no speculation)
Resume prompt
- A compact prompt that can be pasted into a new session

Where to store it

Default: write to
```
outputs/runtime/memory-fold.json
```
(or similar session output)
If user requests: also write a human-readable
```
memory-fold.md
```

Minimal template (copy/paste)

{
  "working_memory": {
    "goal": "",
    "sub_goal": "",
    "blockers": [],
    "next_actions": []
  },
  "tool_memory": {
    "used": [],
    "worked": [],
    "failed": [],
    "availability": []
  },
  "evidence_memory": {
    "anchors": []
  },
  "decision_log": [],
  "resume_prompt": ""
}