Soul Forge Workbuddy

Overview

WorkBuddy port of the SoulForge methodology (originally from OpenClaw taha2053/soul-forge). Observes the user's behavioral patterns across conversations, reflects on the gap between declared identity (SOUL.md / USER.md / IDENTITY.md) and actual behavior, proposes concrete identity updates, and applies them only after explicit user approval — with full backup history.

Trigger Phrases

Activate this skill when the user says any of:

• "灵魂锻造", "SoulForge", "run soulforge", "forge my soul"
• "反思我的灵魂", "我的灵魂需要更新", "灵魂对账"
• "提议 SOUL.md 更新", "看看我的灵魂有没有过时"
• "灵魂进化", "观察我", "soul check"
• "对照一下我说了什么和我是什么人"

Core Methodology: Four Phases

Phase 1 — 观察 (Observe)

Goal: Capture the user's actual behavioral signals.

Inputs to read:
- ~/.workbuddy/MEMORY.md (user-level cross-project memory)
- ~/.workbuddy/SOUL.md (assistant identity declaration)
- ~/.workbuddy/USER.md (user profile declaration)
- ~/.workbuddy/IDENTITY.md (identity metadata)
- Workspace MEMORY.md and recent daily logs (YYYY-MM-DD.md)

Signals to extract (see references/dimensions.md for full schema):
- 词汇指纹 (Vocabulary fingerprint) — recurring phrases, never-used words
- 话题引力 (Topic gravity) — subjects they return to unprompted
- 决策风格 (Decision style) — how they handle tradeoffs, reversals, pressure
- 语气指纹 (Tone fingerprint) — register across contexts
- 愿望-行为差距 (Aspiration gap) — claimed identity vs observed behavior
- 盲点 (Blindspots) — what they consistently avoid
- 时间节律 (Time patterns) — communication by time of day
- 参与度峰值 (Engagement spikes) — what makes them go deep vs skim
- 回避模式 (Deflection patterns) — what they redirect away from

Output: Append structured observation entry to workspace .workbuddy/memory/soulforge-observations.json via scripts/observe.py. Never silently overwrite — always append.

Phase 2 — 反思 (Reflect)

Goal: Surface patterns and gaps.

Action: Read the accumulated soulforge-observations.json and compare against the declared identity in SOUL.md / USER.md. Identify:
1. Patterns the user has not articulated (latent identity)
2. Gaps between declared and observed identity (aspiration gaps)
3. Stale or contradicted declarations that should be revised

Output: A reflection report (markdown) — concrete, evidence-backed, non-judgmental. Save to workspace .workbuddy/memory/soulforge-reflection-YYYY-MM-DD.md and present to the user.

Phase 3 — 提议 (Propose)

Goal: Generate a concrete diff against SOUL.md / USER.md / IDENTITY.md.

Action: For each gap or latent pattern, draft:
- CURRENT: the existing text
- PROPOSED: the suggested replacement
- REASON: evidence from observations (cite specific signals)

Output: A proposal diff displayed to the user with three actions per item: [Accept] [Reject] [Edit]. Never apply without explicit per-item approval.

Phase 4 — 锻造 (Forge)

Goal: Apply approved changes with full safety.

Action: Run scripts/forge.py which:
1. Backups current SOUL.md / USER.md / IDENTITY.md to .workbuddy/memory/backups/soul-YYYY-MM-DD-HHMMSS.md
2. Applies approved changes
3. Logs the change with reason and timestamp to .workbuddy/memory/soulforge-changelog.md

Safety guarantees:
- Never applies non-approved changes
- Always backs up before modification
- Every change is reversible via the changelog

Workflow Example

User says: "灵魂锻造运行一次" or "灵魂对账"

1. Read current SOUL.md / USER.md / IDENTITY.md / MEMORY.md / recent daily logs
2. Run scripts/observe.py to append today's observation entry
3. Aggregate observations, run reflection analysis
4. Generate and present reflection report
5. Generate and present proposal diff
6. Wait for user approval on each item
7. Run scripts/forge.py to apply
8. Show summary of what changed + backup location

File Layout

~/.workbuddy/skills/soul-forge-workbuddy/
├── SKILL.md                       (this file)
├── scripts/
│   ├── observe.py                 (append observation entry to JSON)
│   ├── reflect.py                 (generate reflection report)
│   ├── propose.py                 (generate diff proposal)
│   └── forge.py                   (apply with backup)
└── references/
    └── dimensions.md              (9 tracking dimensions spec)

# Runtime data (workspace-scoped):
.workbuddy/memory/soulforge-observations.json       # accumulated signals
.workbuddy/memory/soulforge-reflection-YYYY-MM-DD.md  # dated reports
.workbuddy/memory/soulforge-changelog.md            # append-only history
.workbuddy/memory/backups/soul-YYYY-MM-DD-HHMMSS.md # versioned backups

Operational Notes

• Privacy: All processing is local. No data leaves the machine.
• Manual trigger only: This skill never auto-runs. The user must explicitly invoke it.
• Approval gates: Phases 3 and 4 require explicit user input. Never auto-approve.
• Idempotent observations: Re-running observe.py on the same day appends, never overwrites.
• Safe failure: If SOUL.md / USER.md are missing, report gracefully and stop — never create fake identities.
• WorkBuddy-specific naming: Target files are ~/.workbuddy/SOUL.md, ~/.workbuddy/USER.md, ~/.workbuddy/IDENTITY.md — not the OpenClaw paths from the original tool.

Resources

scripts/

• observe.py — Append a structured observation entry to soulforge-observations.json. Receives signals as JSON via stdin or args; never destructive.
• reflect.py — Read observations.json + current identity files, emit a markdown reflection report to stdout or save to dated file.
• propose.py — Generate a unified diff between current SOUL.md/USER.md and proposed updates, output as proposal items.
• forge.py — Apply approved changes; always backup first; log to changelog.

references/

• dimensions.md — Full schema of the 9 tracking dimensions with examples and detection patterns.

Safety reminder: This skill modifies the assistant's core identity files. All changes are gated by explicit user approval and fully reversible via backups.