Exécutez n'importe quel Skill dans Manus
en un clic

Exécutez n'importe quel Skill dans Manus en un clic

memory-manager

Étoiles1

Forks0

Mis à jour1 juin 2026 à 21:32

Persistent memory layer across sessions. Use to recall prior context at the start of work, to store checkpoints at phase boundaries, and to answer "what did we decide / do before" questions. Backed by the agentmemory server (local SQLite + vector embeddings) on port 3111, with an MCP-only fallback. Every phase skill calls this to survive /clear and compaction boundaries.

Installation

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Exécuter dans Manus

Source

AZANIR

AZANIR/sdlc-kit

Ouvrir le dépôt GitHub Voir les dépôts du créateur

Téléchargement

Exécuter dans Manus

Métiers associésSOC

Basé sur la classification professionnelle SOC

Développeurs de logicielsProfessions informatiques et mathématiques·SOC 15-1252

SKILL.md

readonly

name	memory-manager
description	Persistent memory layer across sessions. Use to recall prior context at the start of work, to store checkpoints at phase boundaries, and to answer "what did we decide / do before" questions. Backed by the agentmemory server (local SQLite + vector embeddings) on port 3111, with an MCP-only fallback. Every phase skill calls this to survive /clear and compaction boundaries.

Memory Manager (persistent layer)

Adapted from rohitg00/agentmemory. Gives the SDLC recall across sessions so work resumes instead of restarting.

Architecture

Server — local agentmemory server on http://127.0.0.1:3111 (started by the SessionStart hook via scripts/memory-up.*). Uses local SQLite + vector embeddings (@xenova/transformers, no external API).
Fallback — if the native server/iii-engine cannot run (esp. Windows), use the MCP server npx -y @agentmemory/mcp. Recall quality is lower but the workflow still functions.
Config — AGENTMEMORY_INJECT_CONTEXT=true and AGENTMEMORY_AUTO_COMPRESS=true are set in settings.json (both ship OFF by default upstream — we turn them ON).

Operations

recall(project, query?) — GET /recall. Do this at session start and before big decisions.
store(record) — POST /store. Do this at every phase boundary with keys in English: {project, kind, phase, artifact, decision, next, note}. Values may be Ukrainian.
checkpoint — a store with kind="checkpoint" and a "RESUME HERE" note.

When to call

SessionStart (hook does it) — recall.
End of each phase — store checkpoint.
Before /clear or on PreCompact (hook does it) — store checkpoint.
When the user asks what was decided/done previously — recall + answer.

Degradation

If memory is unavailable, say so explicitly and continue using .sdlc/state.json + the phase artifacts on disk as the fallback source of truth. Never block the workflow on memory.

Примітка (UK)

Шар постійної пам'яті між сесіями: локальний сервер agentmemory (SQLite + векторні ембединги) на порту 3111, з MCP як резервом. Ключі записів — англійською, значення можуть бути українською. Якщо памʼять недоступна — працюємо за файлами на диску й попереджаємо.

Plus depuis ce dépôt

même dépôt

code-graph

AZANIR/sdlc-kit

Optional code-navigation aid for large or unfamiliar codebases. Use during the plan or implement phases to map module dependencies, find where a symbol is defined and used, and orient before making changes. This is a navigation helper only — it is NOT the memory layer. Skip for small projects.

2026-06-011

sdlc-design

AZANIR/sdlc-kit

Phase 4 of the SDLC, used ONLY for user-facing features. Use this to produce UI/UX design documentation and a persistent design system before implementation. Triggers when tasks.md exists and the spec marked a UI surface. Coordinates with the ui-ux-pro-max skill to generate design-system/MASTER.md and per-page specs. Skip entirely for CLI tools, libraries, or backend-only work.

2026-06-011

sdlc-implement

AZANIR/sdlc-kit

Phase 5 of the SDLC. Use this to write code that satisfies the tasks, strictly test-first (TDD). Triggers when tasks.md exists and design is done or not needed. Implements one task at a time, writing a failing test before code, then making it pass, then refactoring. Persists progress to memory after each task so work survives session boundaries.

2026-06-011

sdlc-orchestrator

AZANIR/sdlc-kit

Master coordinator for the full software development lifecycle. Use this when starting a new feature or project, when the user is unsure which phase to run, when resuming work after a break, or to establish/edit the project constitution. Routes to specify, plan, tasks, design, implement, review, and ship phases in the correct order and enforces that no phase is skipped. ALWAYS consider this skill first at the start of any development work.

2026-06-011

sdlc-plan

AZANIR/sdlc-kit

Phase 2 of the SDLC. Use this to decide HOW a specified feature will be built — architecture, tech choices, data model, API contracts, and trade-offs. Triggers after a spec exists and the user wants a technical plan or design decisions. Produces specs/<feature>/plan.md plus Architecture Decision Records (ADRs). Must read spec.md and memory/constitution.md and respect their constraints.

2026-06-011

sdlc-review

AZANIR/sdlc-kit

Phase 6 of the SDLC. Use this to verify the implementation against the spec's acceptance criteria and the project constitution before shipping. Triggers when implementation is complete. Produces specs/<feature>/review.md — a filled acceptance checklist with pass/fail per criterion and required fixes. Does not ship; gates whether ship is allowed.

2026-06-011

name	memory-manager
description	Persistent memory layer across sessions. Use to recall prior context at the start of work, to store checkpoints at phase boundaries, and to answer "what did we decide / do before" questions. Backed by the agentmemory server (local SQLite + vector embeddings) on port 3111, with an MCP-only fallback. Every phase skill calls this to survive /clear and compaction boundaries.

Memory Manager (persistent layer)

Adapted from rohitg00/agentmemory. Gives the SDLC recall across sessions so work resumes instead of restarting.

Architecture

Server — local agentmemory server on http://127.0.0.1:3111 (started by the SessionStart hook via scripts/memory-up.*). Uses local SQLite + vector embeddings (@xenova/transformers, no external API).
Fallback — if the native server/iii-engine cannot run (esp. Windows), use the MCP server npx -y @agentmemory/mcp. Recall quality is lower but the workflow still functions.
Config — AGENTMEMORY_INJECT_CONTEXT=true and AGENTMEMORY_AUTO_COMPRESS=true are set in settings.json (both ship OFF by default upstream — we turn them ON).

Operations

recall(project, query?) — GET /recall. Do this at session start and before big decisions.
store(record) — POST /store. Do this at every phase boundary with keys in English: {project, kind, phase, artifact, decision, next, note}. Values may be Ukrainian.
checkpoint — a store with kind="checkpoint" and a "RESUME HERE" note.

When to call

SessionStart (hook does it) — recall.
End of each phase — store checkpoint.
Before /clear or on PreCompact (hook does it) — store checkpoint.
When the user asks what was decided/done previously — recall + answer.

Degradation

If memory is unavailable, say so explicitly and continue using .sdlc/state.json + the phase artifacts on disk as the fallback source of truth. Never block the workflow on memory.