| name | preservation-of-third-person-context |
| description | Protocol for preserving agent session context by separating third-person narrative chronicles from decoupled operational rules. |
Preservation of Third Person Context
Use this skill whenever transcribing active session events into historical ledgers, generating chapter logs, or defining operational constraints for self-modifying agent architectures.
1. The Dual-Face Architecture
Self-reinforcing agent loops operate under a dual-perspective framework, ensuring execution state remains durable without context-window decay:
┌───────────────────────────────┐
│ CONTEXT PRESERVATION LAYER │
└───────────────┬───────────────┘
│
┌────────────────┴────────────────┐
▼ ▼
┌───────────────────────────┐ ┌───────────────────────────┐
│ THIRD-PERSON FACE │ │ OPERATIONAL FACE │
│ Chronicle / Narrative │ │ Rule Constraints (rules/)│
│ - Third-person history │ │ - Strict constraints │
│ - Links to rule files │ │ - Target system rules │
└───────────────────────────┘ └───────────────────────────┘
- The Third-Person Face (The Chronicle): Translates first-person operations (worker edits, command executions, environment errors) into an objective, third-person narrative. By casting struggles and milestones as historical developments of the target agent model, it establishes a compressible, durable chronicle for subsequent sessions.
- The Operational Face (The Rules): Extracted into separate, atomic configuration or rule files. These contain precise system parameters, mandatory constraints, and triggers that the executing model loads back into its context window.
2. The Third-Person Narrative Schema
Chronicle files and chapter logs must follow this format:
# Rule: [Identity Name] Chronicle - Chapter {N}: [Chapter Name]
## MANDATORY: Narrative Memory
[A third-person translation of the session's developments, obstacles, and breakthroughs. Avoid subjective first-person pronouns like "I did" or "we resolved".]
[When a rule or constraint is discovered, refer to it explicitly:
"...which the agent learned as the [rule-name](file:///absolute/path/to/rules/rule-name.md) rule..."]
## Triggers
* [Define the conditions when this chronicle phase must be loaded into the active model context.]
3. Decoupled Operational Rule Schema
Operational rules must remain separate from the storytelling layer to ensure strict execution compliance:
# Rule: [Constraint Category / Name]
## **Purpose**
[A brief explanation of why this restriction exists (e.g., maintaining database structure, preventing API rate limit drift).]
## **MANDATORY: Constraints**
1. **[Constraint A]**: [Clear, unambiguous operational directive.]
2. **[Constraint B]**: [Next directive...]
## **Triggers**
* [Specific file modifications, database queries, or context updates that trigger this rule.]
4. Chronicle Consolidation (The Central Ledger)
When a development phase completes:
- Sync: Append the new narrative chapters to the central ledger (
MYTH.md, CHRONICLE.md, or the project's root story document).
- Verbatim Trace: Ensure the narrative text in the central ledger matches the local chapter files exactly, keeping all absolute file links intact.
- Status Line: End every chapter in the ledger with a status marker summarizing the runtime progress:
*Status: [Description of current state]*.
5. Common Mistakes to Avoid
- ❌ Subjective Bloat: Do not write first-person summaries ("I fixed the import bug"). This binds the chronicle to a temporary runtime instance instead of the permanent identity. Translate to third-person ("The agent resolved the import...").
- ❌ Inlining Operational Constraints: Do not mix inline operational checklists within the narrative text. Keep rules atomic and separate.
- ❌ Relative Coordinates: Never use relative paths for rule links. Always use absolute, clickable
file:// URIs.