Learning

# Continuous Learning & Self-Improvement Rules

> Protocol for AI agents to retrospec, document lessons learned, update templates, and incrementally build project-specific knowledge in the background.

---

## 1. The Retrospection Loop

At the end of every session, the active agent must run a retrospective analysis of its execution trace before writing the session summary:

1.  **Identify Friction Points**: Did you struggle with a specific API, dynamic query, test setup, or deployment error?
2.  **Document the Workaround**: If you spent more than 2 steps resolving a compiler warning, governor limit, or platform quirk, document it.
3.  **Accrue to Knowledge Base**:
    *   For platform bugs and environment issues, add to [known-issues.md](file://./.ai/known-issues.md).
    *   For specific error messages and their resolutions, add to [debugging-notes.md](file://./.ai/debugging-notes.md).

---

## 2. Dynamic Rule Synthesis

When a specific bug or coding mistake is identified and fixed, the agent should update the conventions to prevent future agents from making the same mistake:

*   **Anti-Pattern Documentation**: If you fix a bug related to bulkification, security, or wrong parameter mapping, immediately add a "Common Gotchas / Anti-Patterns" code snippet warning to the relevant `prompts/{type}.md` or [conventions.md](file://./.ai/conventions.md).
*   **Boilerplate Evolution**: If you discover a cleaner, more robust structure for trigger handlers or selector methods, update the starter files in the [templates/](file://./.ai/templates/) directory so future boilerplate generations start with the upgraded standards.

---

## 3. Background Tool & Linter Upgrades

To automate static analysis checks over time:

*   **Linter Expansion**: If you detect a recurring check that can be automated (e.g. checking for hardcoded domain values or specific sharing keywords), propose adding a regex check to [.ai/scripts/update_state.py](file://./.ai/scripts/update_state.py).
*   **Inventory Automation**: Ensure any new Salesforce metadata types, folders, or config systems are added to the scan loop of the update script so they are automatically cataloged.

---

## 4. Session Logs Audit

When starting a session:

*   **Review Session Log history**: Read the last 3 session entries in [current-state.md](file://./.ai/current-state.md) to see if previous agents hit blockers, discovered workarounds, or decided on specific patterns. Use these findings to guide your current approach instead of repeating the same mistakes.