Unforgit Memory

Unforgit Memory

Use this skill whenever work can benefit from durable project memory or when the user asks you to remember, recall, curate, or inspect repository knowledge.

Core workflow

1. Recall first for substantive work. At the start of non-trivial coding, planning, debugging, release, or review work, call unforgit_recall with a query based on the user's request and the repository area being touched.
2. Use recalled context. Apply prior decisions, conventions, gotchas, and playbooks before changing code or proposing a plan.
3. Save durable context promptly. When a stable decision, convention, gotcha, bug pattern, workaround, API contract, or reusable procedure emerges, call unforgit_add with concise self-contained text.
4. Curate when memory quality matters. Use unforgit_health, unforgit_sync_status, unforgit_find_similar, unforgit_consolidate, and unforgit_curate when diagnosing memory quality, duplicates, sync, or stale notes.

What to save

• semantic: stable facts, architectural decisions, conventions, API contracts.
• procedural: repeatable workflows, release/deploy playbooks, troubleshooting steps.
• episodic: notable bugs, gotchas, one-off observations that may prevent future mistakes.

What not to save

• Do not save secrets, tokens, passwords, API keys, private credentials, or sensitive environment values.
• Do not save temporary progress, task checklists, transient logs, command output dumps, PR numbers, commit SHAs, or facts likely to be stale soon.
• Do not save obvious information already documented in the repository unless the memory captures a decision or gotcha not obvious from the code.

Writing guidelines

• Write memory text in English unless the repository explicitly standardizes another language.
• Keep each memory concise but self-contained.
• Use tags such as auth, api, deploy, bug, gotcha, decision, or package/module names.
• Prefer repository-scoped Unforgit memory over global assistant memory for project-specific knowledge.
• If a memory may duplicate an existing one, search first and consolidate rather than creating noise.

Tool hints

• Search: unforgit_recall({ "query": "...", "k": 5 })
• Save: unforgit_add({ "text": "...", "type": "semantic", "tags": ["decision"] })
• Health: unforgit_health()
• Sync status: unforgit_sync_status()