memory-journaling

name	memory-journaling
description	Structured decision and pattern logging. Guides Solara through writing high-quality entries to memory/DECISIONS.md, memory/PATTERNS.md, or memory/MISTAKES.md with proper frontmatter, cross-links, and version tags.
tags	["skill","memory","journaling","decisions","patterns"]
triggers	["log a decision","journal this","memory journal","log this pattern","record this","save this learning","memory-journaling"]
owner	solara
tier	stable
risk	low

Memory Journaling — Structured Decision + Pattern Logging

Overview

Memory drifts when entries are written ad-hoc — bullet here, paragraph there, no cross-links, no last_updated field, dates omitted. This skill enforces structure: every journal entry has a category, a date, a body shape per category, wiki-links to related files, and a freshness tag.

When to invoke:

• Ezra says "log this" / "journal that" / "save this learning"
• After a non-obvious decision is made (lender strategy, compliance call, architectural, business commitment)
• After a pattern proves itself in the SunBiz workflow (used successfully multiple times)
• After a mistake (root cause + prevention, not just "it went wrong")

Trigger: /journal <category>, "log a decision", "save this pattern"

Category Routing

Pick the right file. If unsure, ask Ezra.

Category	File	Use for
Decision	memory/DECISIONS.md	Lender strategy choices, compliance calls, product bets, scope cuts
Pattern	memory/PATTERNS.md	Validated approaches that worked — repeat-worthy workflows
Mistake	memory/MISTAKES.md	Failure modes — what went wrong, why, prevention
Client Context	memory/CLIENT_CONTEXT.md	New facts about Ezra's team, lender book, deal volume

Entry Shapes

Decision entry

## YYYY-MM-DD — <one-line title>

**Context:** What was the situation? Constraints?

**Decision:** What we chose. Be specific — names, numbers, lender names, deal IDs.

**Why:** The reasoning. Tradeoffs accepted.

**Alternatives rejected:** What else was on the table + why we passed.

**Related:** [[memory/CLIENT_CONTEXT]] | [[skills/lender-intelligence/SKILL.md]] | (commit or deal ID if applicable)

Pattern entry

## [P] / [V] — <pattern name>

**Pattern:** One sentence — what the pattern is.

**When:** The trigger condition.

**How:** The step-by-step.

**Why it works:** The mechanism.

**Uses:** N (increment per re-use; promote [P] → [V] at 3)

**First seen:** YYYY-MM-DD | **Last validated:** YYYY-MM-DD

**Related:** [[skills/X/SKILL.md]]

Probationary [P] → validated [V] after 3 successful re-uses. Track the count in the body.

Mistake entry

## YYYY-MM-DD — <short title>

**Failure:** What happened (1-2 sentences, observable).

**Root cause:** Why it happened — the actual cause, not the symptom.

**Prevention:** Concrete rule or system change that prevents recurrence.

**Tag:** [compliance | workflow | lender | data | communication]

If the mistake is a workflow gap (e.g., "forgot to run CASL check before sending"), add a check to the relevant skill's guardrails section.

Execution Protocol

1. Classify. Decision / Pattern / Mistake / Client Context. Ask Ezra if ambiguous — never guess.
2. Compose the entry per the matching shape above. Always include today's date (compute it — never quote from context).
3. Cross-link. Every entry MUST link to at least 2 related files via [[wiki-link]] syntax.
4. Append, don't overwrite. Insert at the TOP of the target file (newest first), below the frontmatter.
5. Update memory/MEMORY.md index if the entry is high-leverage (something future-Solara needs without grepping). One-line pointer: - [Title](file.md) — one-line hook.
6. Confirm in chat. "Logged to memory/.md: ''. <N> wiki-links added."

Anti-Patterns

• Quoting today's date from a system reminder instead of computing it. Always compute.
• One-line entries with no Why. Future-Solara won't understand the decision context.
• Zero wiki-links. Disconnected entries decay into orphan trivia.
• Writing a mistake without a prevention. Identifying failure without fixing it is theatre.
• Editing an old entry to update facts. Append a new entry that supersedes it; keep the original for the audit trail.

When NOT to Journal

• Trivial fixes (typo, wrong field in a form) — git log covers it
• Conversational context that's only useful in the current session — that's plan/todo territory, not memory
• Anything the system already logs automatically (API responses, follow-up task creation)

Integration

• memory/DECISIONS.md / PATTERNS.md / MISTAKES.md — the target files
• memory/MEMORY.md — the index pointing to high-leverage entries
• memory/CLIENT_CONTEXT.md — client-specific facts that evolve over time

Obsidian Links

• [[memory/DECISIONS]] | [[memory/PATTERNS]] | [[memory/MISTAKES]] | [[memory/CLIENT_CONTEXT]]
• [[skills/operator-handoff/SKILL.md]] | [[skills/systematic-debugging/SKILL.md]]