session
/session
Starts or resumes a work session at the lowest context cost. Reads the state file, logs any new request in the owner's words, reads only what the task touches, then proposes a short plan before any code. Use at the start of every session, after /clear or a compaction, or when the user says "let's resume", "continue", "next", "where are we".
- Author
- guillaume-piron-dev ↗
- Repository
- you-decide ↗
- Licence
- Apache-2.0
- Last repository change
- 07.11.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 07.11.2026, without references to retired models or APIs.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
What it does and when to use it
Starts or resumes a work session at the lowest context cost. Reads the state file, logs any new request in the owner's words, reads only what the task touches, then proposes a short plan before any code. Use at the start of every session, after /clear or a compaction, or when the user says "let's resume", "continue", "next", "where are we".
Triggers from the description
Original author description
Starts or resumes a work session at the lowest context cost. Reads the state file, logs any new request in the owner's words, reads only what the task touches, then proposes a short plan before any code. Use at the start of every session, after /clear or a compaction, or when the user says "let's resume", "continue", "next", "where are we".
How it works
- Steps
- Budget
- Never
SKILL.md preview
--- name: session description: Starts or resumes a work session at the lowest context cost. Reads the state file, logs any new request in the owner's words, reads only what the task touches, then proposes a short plan before any code. Use at the start of every session, after /clear or a compaction, or when the user says "let's resume", "continue", "next", "where are we". --- # Start or resume a session Goal: be operational after reading less than ~10,000 tokens. Never read large logs or registers in full. ## Steps Light level (only CLAUDE.md and `docs/state.md` exist): skip the logs and decision records below; check new requests against CLAUDE.md and `docs/state.md`, and treat the state file's next actions as the plan. 0. After a compaction, the resume context is already in the conversation (hook); keep the memory with the `memory` skill and re-read each file it names before acting. 1. Read `docs/state.md`, the only file read in full. If it is missing, suggest `/you-decide:setup` and stop. To know where to search: `docs/memory/map/README.md` (standard and strict levels). 2. New request from the owner? Log it first, in their exact words (skill `log`). Flag any contradiction with an earlier request, rule or decision (grep the topic in `docs/requests_log.md` and `docs/decisions/`). 3. Read only what the task touches, by targeted search: - the step: grep its id in the plan file named in `docs/state.md`, plus that milestone's deferred items; - the user story: grep its id, read its title and acceptance criteria only; - open findings for this step: grep `open` and the step id in `docs/findings_log.md`; - a decision: grep its id in `docs/decisions/`, read that record. 4. Propose a short plan **before writing code**: files touched, risks, tests, and anything that needs the owner's yes (network, new dependency or service, deletion, irreversible action). One session = one task. 5. Any legal, security or scope doubt: stop and ask, with a multiple-choice question when possible. ## Budget - Use the cheapest model that is good enough for routine steps. Switch to the most capable one for architecture decisions or a bug the first one could not solve. - `/clear` between two tasks. Plan mode before writing code. Subagents only for a wide, isolated read. - To check a fix, run the test that proves it. Do not re-run a full review. ## Never - `/clear`, or let a compaction happen, while a decision of the owner is written nowhere. First log it and update `docs/state.md`: what is not in a file is lost at `/clear` and can be distorted by a compaction summary. - Re-read CLAUDE.md, the logs, the plan and the decisions in full after a compaction. Read `docs/state.md`, search the rest. - Start a long run, a download or any network access without the owner's yes.
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · guillaume-piron-dev/you-decide | 90 | Apache-2.0 |
| glassflow/tares | 100 | MIT |
| r3dlex/oh-my-auggie | 100 | Apache-2.0 |
| lex0tan/backlog-resume | 100 | MIT |
| rdennler/yat | 100 | Apache-2.0 |
| rdennler/thinkpad | 100 | Apache-2.0 |
| maxwell2732/claudecode-research-harness-workflow | 100 | MIT |
| dotroza/hermes-comfy-recipe | 96 | MIT |
| autocss-com/ai | 90 | MIT |
Database snapshot: 13.11.2026 · package version 1b7f67af5d09 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90session-resume
Use when starting a session that should continue prior work — phrases like "resume the session", "pick up where we left off", "continue from where we stopped", "where were we", "load the continuation file". Consumes the .claude/CONTINUE.md written by session-save: verifies it… - 100new-session
Start a new session when context is high. Claude uses /clear, Codex uses /exit. Use when context is high or when a fresh session is needed. - 90safe-resume
Use when the user says "safe resume", "resume from pause", "pick up where we left off", "continue the paused work", or when a session start hook reports a /dev:safe-pause state for this session. Also use when a fresh session is asked to continue and a RESUME.md exists for it. - 100continue-context
Take over the work a context file describes and drive it as the main agent. Use when the user or a resume prompt names a context file (a task context.md, the project context.md) and asks to continue, resume or take over; when a session is woken with no context by a notification;… - 90repack
Refresh a long chat — write the next-session handoff, then hand the user /clear → /continue so the work resumes in a fresh context window. Use when the context nudge reaches the handoff level, or the user says "repack", "refresh the chat", "context is getting full", "fresh… - 100load-context
Resume session from CONTEXT-llm.md. Use when resuming work, loading saved context, continuing a previous session. Triggers include "load context", "resume session", "continue where I left off".