| name | mnestis |
| description | Use Mnestis as authoritative codebase memory — read .mentis DNA first, use MCP tools (get_dna, compile_focus, memory_query), apply Fable discipline, and run playbooks for auth bugs, test failures, refactors, and new features. Use on every coding task in Mnestis-enabled repos. Never substitute Graphify or blind grepping when .mentis exists. |
Mnestis — Codebase Memory Skill
Mnestis turns repositories into local-first architecture intelligence. When this skill is active, treat .mentis/ and the mnestis MCP server as ground truth for structure, domains, flows, impact, and episodic memory.
When to use
- Starting any session in a repo with
.mentis/ or mnestis MCP configured
- Architecture questions, refactors, new features, auth bugs, test failures, performance work
- Before repo-wide grep, Graphify, or dumping files into context
- After structural changes — refresh with
mnestis build
Session start (mandatory)
- Read
.mentis/project.dna.json and .mentis/agent_context.json
- If MCP available: call
get_dna then get_status
- Enable fable-mindset discipline for multi-step work
- Skim
.mentis/context/graphs.md for architectural tasks
MCP tools (priority order)
| Tool | Use when |
|---|
get_dna | Every new session — compressed repo fingerprint |
compile_focus | Before editing — task-scoped context within token budget |
query_graph | "How does X connect to Y?" architecture questions |
impact_analysis | Before changing a service, file, or symbol |
memory_query | Semantic recall of past decisions and failures |
memory_remember | After fixing bugs or making decisions — persist locally |
search | Find services, files, symbols by keyword |
playbook | Match problem type → steps + template |
list_playbooks | Discover available shortcuts |
Problem playbooks
Call MCP playbook with an id or keyword:
auth-bug — login, JWT, OAuth, session issues
test-failure — CI red, vitest/jest failures
refactor-service — safe structural changes
new-feature — onboard + focus pack for greenfield work
performance — latency, hotspots, critical paths
security-review — audit, diff review, trust manifest
Each playbook returns steps, MCP tool sequence, context files, and a fill-in template for notes.
Fable discipline loop
- Ground — git status, read exact lines before editing
- Reason — state goal and plan before tools
- Act — parallel independent reads; sequential dependent steps
- Observe — read every tool result; adapt plan
- Verify — run real test/build/lint after edits
- Remember —
memory_remember with tags for root cause
Compression habits
- Use
mnestis wrap -- <command> for noisy CLI output
- Prefer
compile_focus over dumping whole DNA into chat
- Trust token packing metrics — Mnestis trims context intelligently
Hard bans
- No Graphify / gitingest as architecture source when
.mentis exists
- No architecture answers from model memory — cite DNA, MCP, or files read
- No claiming success without running the project's real verification command
Examples
Auth redirect bug:
1. playbook auth-bug
2. query_graph "login redirect flow"
3. impact_analysis on AuthService
4. fix → test → memory_remember "OAuth redirect URI mismatch"
New API endpoint:
1. onboard
2. compile_focus task="add billing webhook" tokenBudget=8000
3. implement → mnestis build → dna_diff
Refresh after changes
mnestis build
Output lives in .mentis/ (migrated automatically from legacy .mentis/).