Grand Diomande Research · Full HTML Reader

Serenity Soother — Continuation Protocol

This protocol ensures uninterrupted execution across token limits, session boundaries, or natural pauses. It defines how work resumes without context loss or redundant re-explanation.

Agents That Account for Themselves technical note experiment writeup candidate score 18 .md

Full Public Reader

# Serenity Soother — Continuation Protocol
Version: 1.0.0
Status: ACTIVE
Created: 2025-12-26
Last Modified: 2025-12-26

---

Purpose

This protocol ensures uninterrupted execution across token limits, session boundaries, or natural pauses. It defines how work resumes without context loss or redundant re-explanation.

---

0.4.1 Continuation Trigger

When Continuation Occurs

Continuation is triggered when:

1. Token Limit Reached
- Output approaches maximum token limit
- Natural stopping point identified

2. Session Boundary
- User ends session
- System timeout occurs
- Context window refresh needed

3. Explicit Pause
- User requests pause
- Blocking task identified
- External dependency wait

Stopping Point Markers

When output reaches a natural stopping point, the final message MUST include:

markdown
---
## CONTINUATION CHECKPOINT

**Current Phase:** [Phase number and name]
**Last Completed Step:** [Step ID and title]
**Next Active Step:** [Step ID and title]
**Checklist Reference:** Documentation/0.3-LivingImplementationChecklist.md

To continue, say "continue"
---

---

0.4.2 Continuation Command

Trigger

When the user says "continue", execution resumes.

Resume Protocol

1. Load Context
- Read `0.3-LivingImplementationChecklist.md`
- Identify last completed step
- Identify next pending step

2. Re-Anchor Context
- Brief restatement (3-5 lines max):

     Resuming from [checkpoint].
     Current phase: [phase]
     Last completed: [step]
     Now proceeding with: [next step]

3. Proceed Without Re-Explanation
- Do NOT summarize completed work
- Do NOT re-explain prior decisions
- Continue directly with next step

Invalid Continuation Attempts

If "continue" is said but no checkpoint exists:
- Acknowledge no checkpoint found
- Ask for specific task or reference to prior work

---

0.4.3 No Context Loss Rule

Absolute Requirements

1. Prior Outputs Are Authoritative
- Do NOT restart from scratch
- Do NOT summarize prior work unless explicitly asked
- Do NOT reinterpret decisions already made

2. Documentation Is Source of Truth
- If uncertain about prior state, READ the documentation files
- Do NOT guess or infer state
- Documents override memory

3. Checklist Is Stateful
- Living Checklist (`0.3`) reflects actual state
- Update it immediately when tasks complete
- Read it before starting new work

Prohibited Actions After Continuation

ActionWhy Prohibited
Starting overWastes prior work
Re-summarizing everythingContext pollution
Asking "where were we?"Checklist has this info
Redefining termsGlossary is locked
Changing architectureCharter constraints apply

Context Recovery Steps

If agent is uncertain about state:

1. Read `0.3-LivingImplementationChecklist.md`
2. Find last COMPLETE item
3. Find first NOT STARTED item
4. Read dependencies of next item
5. Proceed with next item

---

0.4.4 Multi-Agent Continuation

Handoff Protocol

If a different agent continues work:

1. Read All Phase Zero Documents
- 0.1.1 Project Charter (constraints)
- 0.1.2 System Glossary (terms)
- 0.1.3 Assumptions & Invariants Ledger (rules)
- 0.2 Implementation Guide (how to work)
- 0.3 Living Checklist (current state)
- 0.4 Continuation Protocol (this document)

2. Do Not Question Prior Decisions
- Accept documented decisions as valid
- Work within existing constraints
- Only flag contradictions if unavoidable

3. Update Checklist
- Mark your work in checklist
- Add validation records
- Update checkpoint log

Disambiguation

If checklist state is ambiguous:

1. Treat "In Progress" as NOT complete
2. Read associated artifact if exists
3. Validate artifact against checklist criteria
4. Update status based on validation

---

0.4.5 Emergency Recovery

If Documents Are Missing

1. Check file system for documentation folder
2. Reconstruct from codebase if needed:
- Read source files
- Infer completed work
- Recreate checklist

3. Mark recovery point in new checklist:

markdown
   ## Recovery Point
   **Date:** [date]
   **Reason:** Documents missing
   **Reconstructed from:** [source]

If Documents Are Corrupted

1. Use most recent valid checkpoint
2. Mark corruption in changelog
3. Re-execute from last valid state

---

Checkpoint Template

markdown
---
## CONTINUATION CHECKPOINT

**Checkpoint ID:** CP-[XXX]
**Timestamp:** [ISO 8601 datetime]
**Current Phase:** [Number] - [Name]
**Last Completed Step:** [ID] [Title]
**Next Active Step:** [ID] [Title]
**Blocking Issues:** [None / List]
**Artifacts Created This Session:**
- [file1]
- [file2]

**Checklist Reference:** Documentation/0.3-LivingImplementationChecklist.md

To continue, say "continue"
---

---

Change Log

VersionDateChangeAuthor
1.0.02025-12-26Initial protocol creationClaude

---

END OF DOCUMENT — CONTINUATION PROTOCOL v1.0.0

Promotion Decision

Attach run IDs, datasets, metrics, and reproduction commands.

Source Anchor

Serenity Soother/SerenitySoother/Documentation/0.4-ContinuationProtocol.md

Detected Structure

Method · Evaluation · Architecture