Memory Ledger Protocol

Keep a structured log of everything learned during a task. Goal: when a new instance takes over, it reads the ledger and continues from where things left off — no starting from scratch.

Ledger Structure

One ledger block per task. Updated incrementally as the task progresses — never rewritten from scratch.

MEMORY LEDGER
Task        : [task name / short description]
Started     : [first step number or timestamp]
Last updated: [step number or description]

## Goal
[What is being done — 1-2 sentences. Scope and success criteria.]

## User Preferences & Constraints
- [Preference / constraint 1]
- [Preference / constraint 2]
[If empty: "None established yet"]

## Decisions Made
| # | Decision | Rationale | Step |
|---|----------|-----------|------|
| 1 | [what was decided] | [why] | [which step] |

## Current Status
Completed  : [list or "—"]
In progress: [list or "—"]
Pending    : [list or "—"]

## Failed Approaches
| Approach | Why It Failed | Step |
|----------|---------------|------|
| [what was tried] | [error / issue] | [which step] |
[If empty: "None yet"]

## Bugs & Fixes
| Bug | Fix | Status |
|-----|-----|--------|
| [error message / description] | [applied fix] | Fixed / Open |
[If empty: "None yet"]

## Next Step
[The next thing to do — clear and actionable]

Update Rules

Update the ledger when any of these events occur:

Updates must be small and surgical — only change the affected section, do not rewrite the entire ledger.

Handover Protocol

When a new instance takes over a task, first action is to read the ledger:

LEDGER READ — Task resumed
Task    : [task name]
Status  : [current status summary]
Next    : [next step]
Warning : [if any open bugs or failed approaches]
Continuing...

Ledger Close

When the task is complete, close the ledger:

LEDGER CLOSED
Task completed
Total steps   : N
Decisions made: N
Bugs resolved : N

When to Skip

• Single-step, short tasks (< 5 minutes, < 3 steps)
• User is only asking a question, no task
• Context is already small and manageable

Guardrails

• Never let the ledger grow stale — if 3+ steps pass without an update, force one.
• Keep entries concise — one line per decision, one line per bug. No paragraphs.
• Cross-skill: integrates with task-decomposer (tracks subtask progress), error-recovery (logs errors and fixes), and agent-reviewer (provides retrospective data).

Session Handoff Integration

After 3+ significant updates to the ledger (each completion of a subtask), evaluate whether a session handoff should be suggested.

Suggest handoff when:

• Task involves 5+ file edits since session started
• Multiple architecture decisions were made
• Task is becoming long-running (context filling up)
• User explicitly mentions context concern

Suggestion prompt:

"We've made substantial progress on this task. A session handoff document would preserve this context for future sessions, especially if we need to resume later or if context gets full. Say 'create handoff' when ready to save state."

What to include when creating handoff from memory-ledger:

• Decisions Made table → Decisions Made section
• Current Status → Current State Summary + Pending Work
• Bugs & Fixes → Potential Gotchas
• User Preferences → Assumptions Made
• Next Step → Immediate Next Steps (first item)

For long tasks, chain handoffs: invoke session-transfer skill with --continues-from to link related handoffs.

References

• references/EXAMPLES.md — a live ledger maintained across a multi-step Python project. Read it when the expected output format or level of detail is unclear.

🔗 Next Steps & Handoffs

Ecosystem: @ecosystem-orchestration — Agent Orchestration.

Workflows:

• Complex Task Execution Flow (orch-complex-task, step 4 of 6): next → @output-critic (optional).
• Knowledge Capture Flow (orch-knowledge, step 1 of 2): next → @mastermind-logger.

Direct handoffs:

• @session-transfer — The task will continue in a new session.

Handoff contract: pass results to the next skill through an inline Handoff block in your reply with the fields skill, workflow, created_at, inputs, summary, and next (the handoff contract of @ecosystem-orchestration). If a next skill is not installed, continue with its step from the ecosystem map, or install it with npx skills add fatih-developer/fth-skills --skill <name> after the user agrees.