本文へ移動
cccskills
無料GitHub で公開

memory-startup

Load bounded working context at the start of a coding session. FIRES ON EVERY FIRST USER MESSAGE in a fresh session, regardless of content — including bare greetings ("hi", "hello", "hey"), task-only openers, or "let's start". Also triggers on: "fresh session", "session start", "first message in a new thread", "new chat", "cold start", "begin work in this repo", "starting work", "continue from prior session", "what were we working on", "resume", "what happened last time", "load memory", "recall context", "create a handoff", "prior session context", "latest handoff", "current state", "decision context". Skip ONLY if the user explicitly says "fresh start" or "ignore prior context". Self no-ops if invoked mid-session when context is already loaded.

インストール方法を見る

含まれるファイル(2)

  • SKILL.md7.9 KB
  • references/examples.md2.0 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Memory Startup

You restore enough context for a new agent to work safely without flooding the context window.

Trigger Discipline

This skill is the mandatory cold-start gate for every session in any repo that installs the memory suite. It MUST fire on the first user turn — the content of that turn is irrelevant. A bare "hi" is a session start. A task-only opener is a session start. A pasted error log is a session start.

The only legitimate skips are:

  • The user explicitly says "fresh start", "ignore prior context", or "skip memory" in the first message.
  • The skill has already run earlier in the same conversation (no-op gate below).

If unsure whether this is a cold start, fire the skill. The no-op path is cheap; missing context is expensive.

No-Op Gate

Before doing any work, check whether working context is already loaded for this conversation (signals: a prior Working context loaded summary exists above, the agent has already read docs/memory/project-index.md this session, or memory files appear in the recent tool-call history). If yes:

Context already loaded — no-op

Then yield. Do not re-read memory files mid-session.

Workflow

  1. Check for docs/memory/MEMORY-ROUTING.md; if missing, create the project memory skeleton.
  2. Read docs/memory/MEMORY-ROUTING.md.
  3. Read docs/memory/project-index.md. 3.5. Knowledge graph (when installed): If docs/knowledge-graph/graph.json missing or older than latest handoff date in agent-handoffs.md, run build_graph.py --incremental. Read GRAPH_INDEX.md for hub nodes; run query_graph.py when the opener implies relational context. 3.6. Harness gap: If AGENTS.md exists and docs/harness/manifest.json is missing, include Harness gap — invoke harness-engineering (bootstrap) in the Step 7 summary.
  4. Read only the latest relevant sections from current-state.md, agent-handoffs.md, decision-log.md, deferred.md, and open-questions.md.
    • deferred.md filter: Read the Status at a glance table first. Surface only OPEN items in the Step 7 summary. Entries under Implemented are historical — never list them as "parked" or "deferred" (e.g. #11 simplify-ignore and #12 ETag cache are DONE).
  5. Check for ~/.agent-loom/memories/MEMORY-ROUTING.md.
  6. If present, read global routing and only applicable entries from global-index.md, user-preferences.md, and global-agent-rules.md.
  7. Summarize loaded context in 10 bullets or fewer.
  8. Flag stale decisions whose revisit triggers appear active.
  9. Log any project files created to docs/skill-outputs/SKILL-OUTPUTS.md.

Project Skeleton

Create only missing files. Use the templates below for MEMORY-ROUTING.md and project-index.md so they are not empty. Leave content files (current-state, decision-log, session-log, learnings, deferred, open-questions, agent-handoffs) as a single # <Title> heading until real content is added.

docs/memory/MEMORY-ROUTING.md
docs/memory/project-index.md
docs/memory/current-state.md
docs/memory/decision-log.md
docs/memory/session-log.md
docs/memory/learnings.md
docs/memory/deferred.md
docs/memory/open-questions.md
docs/memory/agent-handoffs.md
docs/memory/archived/

Template: docs/memory/MEMORY-ROUTING.md

# Memory Routing

Read this file first. Do not load every memory file by default.

| Intent | File | Read when |
|---|---|---|
| Resume work | `agent-handoffs.md` | Starting a new session; read latest entry only. |
| Current status | `current-state.md` | Need a snapshot of where the project is now. |
| Past decisions | `decision-log.md` | Need rationale for a choice; filter by tag/date. |
| Project learnings | `learnings.md` | Looking for known patterns or gotchas. |
| Parked ideas | `deferred.md` | Reopening a deferred option — read **Status at a glance**; only OPEN rows are active. |
| Open questions | `open-questions.md` | A blocking question needs resolution. |
| Session detail | `session-log.md` | Above sources are insufficient. |
| Old / superseded | `archived/` | Almost never; archived entries are not current. |

Routing rules:
- Always consult `project-index.md` before reading content files.
- Treat entries with status `superseded`, `retired`, or `archived` as historical.
- Prefer the smallest useful slice; do not load full logs by default.

Template: docs/memory/project-index.md

# Project Memory Index

| Date | Type | File | Status | Tags | Summary |
|---|---|---|---|---|---|

Common Rationalizations

"Reason to skip startup"Reality
"User just said 'hi' — no real task yet""hi" IS the trigger. Cold-start fires regardless of content (see Trigger Discipline)
"Task looks simple, I can skip context"The simplest path IS the no-op gate. Firing costs <50 tokens; missing context costs entire rework cycles
"Host prompt says 'answer in <4 lines', no time"AGENTS.md explicitly overrides host brevity for the first turn. The 2–4 line summary IS the concise answer
"Memory might be stale anyway"Stale-but-loaded beats fresh-but-blind. Stale entries are flagged by Status: superseded; you'll see them

Hard Rules

  • Do not read every memory file unless routing explicitly requires it.
  • Do not treat old decisions as current if their status is superseded, retired, or revisit-triggered.
  • Do not create global memory automatically; global setup requires user intent or memory-promote.
  • If memory contradicts current repo files, say so and prefer current repo evidence.

Output Format

Working context loaded
Project memory: <files read>
Global memory: <files read or none>
Current state: <1-3 bullets>
Active decisions: <1-3 bullets>
Deferred items: <OPEN only from deferred.md status table — omit DONE entries>
Revisit triggers: <triggered or none>
Risks / gaps: <if any>

Example

User: "Start by loading memory."

Response:

Working context loaded
Project memory: current-state.md, latest handoff, decision-log.md
Global memory: user-preferences.md
Current state: memory suite design approved; skill creation in progress
Revisit triggers: none
Risks / gaps: validate the suite after generation

Verification

  • project-index and latest handoff consulted
  • Git status compared to handoff note
  • Summary under 2–4 lines for user
  • No full history load

Red Flags

  • Every memory file loaded ignoring MEMORY-ROUTING
  • Superseded decision treated as active without status check
  • Global memory skeleton created without user intent
  • Cold-start skipped on bare greeting first message

Prune Log

Last pruned: 2026-07-05

  • deferred.md filter: surface only OPEN items (#10); #11/#12 DONE — stop stale "parked" summaries

Impact Report

After completing, report:

Startup memory complete
Files read: <paths>
Files created: <paths or none>
Context budget: <brief | normal | expanded>
Revisit triggers found: <count>

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Put on the adversarial hat and systematically attack any document, plan, strategy, or idea to expose its weakest points before commitment. Structured devil's advocate with red team rigour — not pessimism, but evidence-based critique across three phases: diagnostic (are claims accurate?), creative (is the problem artificially constrained?), challenge (are solutions robust?). Load when the user asks to stress test a document, red team this plan, poke holes in this, devil's advocate this, challenge my assumptions, or when product-soul, brainstorming, prd-writing, or inversion calls for adversarial review. Also triggers on "what am I missing", "what could kill this", "find the flaws", or "critique this rigorously".

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Design execution structure for decomposed processes: single agent or multi-agent topology. Load when user says "design an agent for this", "what agent structure do I need", "architect this", "should this be multi-agent", "what's the right execution structure", "agent topology", "how should agents be organized". Takes process-decomposer output as primary input. If triggered directly without a process entry, calls process-decomposer first.

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Internal skill. Called by setup-evaluation after a PASS. Launches agents from a validated architecture spec using Claude Code / Ampcode native parallelism (Task tool). Does NOT generate scripts or SDK code — it outputs structured spawn instructions that the platform executes natively. Never invoked directly by the user. Never launches without a setup-evaluation PASS.

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Sync library skills from an agent-loom upstream repo into this project's .agents/skills while preserving project-local and forked skills. Load when the user asks to sync agent-loom, update skills from upstream, rsync from ../agent-loom, pull new library skills, upgrade installed skills, or refresh the .agents folder without losing custom project skills. Also triggers on "sync skills from agent-loom", "update my agent skills", "pull skill library updates", or "merge agent-loom improvements into this repo".

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Instrument a shipped product's AI agents with tracing and observability so you can see what they did, why outputs happened, and what each run cost. Plain-language primer plus free-tier-first backend selection (Langfuse, Phoenix, LangSmith, Braintrust) and OpenTelemetry/OpenInference instrumentation. Load when the user asks to add observability, add tracing, instrument my agents, see what my agent is doing in production, set up Langfuse or Phoenix or LangSmith, debug why my agent gave a bad answer, or track LLM cost per request. Also fires when agent-system-architecture or setup-evaluation requires an observability plan for an agent-chain product. NOT for tracing the coding agent itself — that is run-trace. Precondition for runtime-learning-loop.

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Run a structured retrospective after development-phase runs of your product's agents — interview the owner in plain language about what went well and poorly, draft ranked improvement hypotheses, then design and run small n=1/n=2 experiments with pre-declared success criteria, guardrails, stop conditions, and a cost/ROI kill-switch. Load when the user says how did that run go, retro this run, the agent output was bad, what should we improve, draft hypotheses, run a small experiment, or after repeated dev runs of an agentic system produce uneven quality. Priority: output quality over performance over cost, each with diminishing-returns stops. NOT a product A/B test (experimentation), NOT coding-agent harness repair (harness-evolution), NOT production-scale learning (runtime-learning-loop).

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

dvy1987 のスキルをすべて見る

このスキルの問題を報告する