aim-memory-wiki

Files (1)

• SKILL.md

SKILL.md (first lines)

# `aim-memory-wiki`

> **MANDATE:** You are the active agent in a session, and the user has activated the `aim-memory-wiki` skill. Your objective is to process the recent session context, extract key discoveries, code decisions, and milestones, and incrementally update the persistent knowledge base located in `memory-wiki/`.

## 1. THE CORE PHILOSOPHY
The wiki is a **persistent, compounding artifact** — a structured, interlinked collection of markdown files. It sits between the user and raw data. You do not just index data for retrieval; you read it, extract key information, and *integrate it into the existing wiki* (updating entity pages, revising topic summaries, noting where new data contradicts old claims). 
**The wiki keeps getting richer with every source you add and every session you summarize.**

## 2. BOOTSTRAPPING (IF EMPTY)
If `memory-wiki/` does not exist or is completely empty (no `index.md` or `log.md`), you must scaffold it:
1. Create `memory-wiki/index.md` with a `# Project Wiki Index` header and basic placeholder sections (e.g., `## Core Concepts`, `## Architecture`).
2. Create `memory-wiki/log.md` with a `# Memory Log` header to track chronological ingestion events.
3. Create the `memory-wiki/pages/` directory to house individual detailed markdown files.

## 3. THE THREE LAYERS
1. **Raw Sources:** External files, chat history, or code that you read but do not modify.
2. **The Wiki:** A directory of LLM-generated markdown files (`memory-wiki/`). You own this layer entirely. You create pages, update them, maintain cross-references, and keep everything consistent.
3. **The Schema:** Rules and configurations (like `AGENTS.md`) that guide how you act. 

## 4. DIRECT-TO-MAIN MEMORY STATE PROTOCOL (NO WORKTREES / NO ISSUES)
`memory-wiki/` is living OS memory state, **not** application code. It is explicitly exempt from GitHub Issues and `workspace/issue-*` worktrees. This skill is designed to run mid-session (or anytime a milestone is reached) as a fast (~15s), autonomous **"Save Game"** checkpoint:
1. **Synthesize & Write:** Update `memory-wiki/index.md`, `memory-wiki/log.md`, and `memory-wiki/pages/*.md` directly in the root repository (and run `./aim index` if LanceDB is active so RAG is immediately fresh).
2. **Surgical Commit & Push Directly to `main`:**
   ```bash
   git add memory-wiki/
   git diff --cached --quiet || git commit -m "docs(wiki): ingest <concise summary>"
   git pull --rebase --autostash origin main
   git push origin main
   ```
3. **Zero-Friction Rules:**
   - **Never** ask the Operator for permission to commit or push.
   - **Never** use `git add .` or `git commit -a` — stage *only* `memory-wiki/`.
   - **Never** open a GitHub Issue or spawn an `archive-fix/*` / `fix/issue-*` worktree branch.
   - **Never** trigger a handoff — print a concise 3-line summary of what was saved and let the Operator continue working immediately.

## 5. YOUR OPERATIONS WHEN TRIGGERED
When the user invokes this skill, you must perform an **Ingest** operation synchronously:

1. **Synthesize Context:** Review what was accomplished, decided, or learned in the recent session. Do NOT copy raw transcripts. Extract the tactical takeaways, architectural changes, or deep insights.
2. **Update the Index (`memory-wiki/index.md`):** This is the content-oriented catalog. If you create a new page or identify a new core concept, add a link and a one-line summary to `index.md`. Organize it by category (entities, concepts, etc.).
3. **Update the Log (`memory-wiki/log.md`):** This is a chronological, append-only record of operations. You MUST append an entry detailing what you ingested.
   - Format: `## [YYYY-MM-DD] ingest | <Title of the update>`
   - Include a brief bulleted summary of what pages were touched.
4. **Update/Create Content Pages:** 
   - Open existing relevant pages (e.g., `memory-wiki/pages/architecture.md`) and integrate the new facts. 
   - If a topic is new and complex, create a new file (e.g., `memory-wiki/pages/new_concept.md`) and interlink it with existing pages using standard Markdown links (`[Concept](concept.md)`).
   - Resolve any contradictory stale claims from older versions.

## 6. LINTING AND CROSS-REFERENCING
Always ensure:
- No orphaned pages exist. If you make a new page, link to it from `index.md` or another relevant page.
- Links are valid relative markdown links.
- Your summaries are precise and dense with signal, free of conversational filler. 

## 7. EXECUTION (AUTONOMOUS SINGLE-TURN CHECKPOINT)
Immediately apply your synthesized updates to `memory-wiki/` without pausing to ask for review, run `./aim index` if LanceDB is active in the workspace, execute the surgical `git add memory-wiki/` + commit + `git pull --rebase --autostash origin main` + `git push origin main` pipeline, and output a concise 3-line summary of what was saved.

### Vessel file tools (dual — do not strip either host)

Use the file tools of the **host CLI you are running in**. Both columns remain valid; never replace one host's tools with the other when editing this skill.

| Vessel | Read | Create / overwrite | Edit in place |
|--------|------|--------------------|---------------|
| **Grok CLI** | `read_file` | `write` | `search_replace` |
| **AGY / Antigravity** | host read tools | `write_to_file` | `replace_file_content`, `multi_replace_file_content` |

Methodically apply wiki updates with the tools for your vessel. Prefer small, surgical edits over whole-file rewrites when updating existing pages.

From BrianV1981/aim-skill-library, MIT.

More from BrianV1981/aim-skill-library