start
/start
Use at the START of every session, before your first reply — short sessions too — and again at the wrap or handoff. Checks that strain (the session-strain meter) has signed this session in, signs it in if not, asks the user for permission the right way, and offers once to add this step to the agent's instruction file. Also use whenever strain says a session is not signed. Triggers on: session start, new session, first reply, sign in, strain not signed, strain start, wrap up, handoff, end of session. Also use before dispatching a sub-agent: strain does not apply inside one.
- Author
- jovesun-lab ↗
- Repository
- whetstone ↗
- Licence
- MIT
- Last repository change
- 19.10.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 19.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Use at the START of every session, before your first reply — short sessions too — and again at the wrap or handoff. Checks that strain (the session-strain meter) has signed this session in, signs it in if not, asks the user for permission the right way, and offers once to add this step to the agent's instruction file. Also use whenever strain says a session is not signed. Triggers on: session start, new session, first reply, sign in, strain not signed, strain start, wrap up, handoff, end of session. Also use before dispatching a sub-agent: strain does not apply inside one.
Original author description
Use at the START of every session, before your first reply — short sessions too — and again at the wrap or handoff. Checks that strain (the session-strain meter) has signed this session in, signs it in if not, asks the user for permission the right way, and offers once to add this step to the agent's instruction file. Also use whenever strain says a session is not signed. Triggers on: session start, new session, first reply, sign in, strain not signed, strain start, wrap up, handoff, end of session. Also use before dispatching a sub-agent: strain does not apply inside one.
How it works
- At the start
- At the wrap
- Sub-agents and unattended runs
- Limits the host controls
SKILL.md preview
---
name: start
description: >-
Use at the START of every session, before your first reply — short sessions too — and
again at the wrap or handoff. Checks that strain (the session-strain meter) has signed
this session in, signs it in if not, asks the user for permission the right way, and
offers once to add this step to the agent's instruction file. Also use whenever strain
says a session is not signed. Triggers on: session start, new session, first reply,
sign in, strain not signed, strain start, wrap up, handoff, end of session. Also use
before dispatching a sub-agent: strain does not apply inside one.
---
# Strain — start and wrap
Strain counts how loaded a session has become, but only a **signed** session joins the
chain: the next session of the same agent can then report how this one went. Signing is
one command. These steps make sure it happens at the start, and that the user is asked
properly when the host needs their approval.
Every command below is printed, filled in, in strain's start line (it begins "STRAIN
START") and in every strain tick. Copy it from there — it carries this session's key, the
state folder and the plugin's path.
## At the start
1. **Check.** If strain's start line says "this session is not signed in yet", go on.
Without a start line, run `strain-sign.sh --receipt`: it says "not signed" or
"signed as <name>". Signed → you are done; go to step 6.
2. **Sign.** Run the sign command from the start line. Keep the default name it shows
(`<host> · <project folder>`, the same every session, so sessions link up); add
`--agent '<name>'` only if the user gave you a name.
3. **Describe it for the user.** When you run it, describe the command in the user's
language, in plain words, saying what it does for them — for example "Turn on strain:
track how full this conversation is". Your host shows that sentence when it asks the
user to approve the command; it is all they see.
4. **If your host asks the user to approve it,** keep the prompt open until they answer —
do not end your turn. If your host has no prompt, ask in chat, then end your turn and
wait for the reply.
5. **Allowed:** tell the user the one sentence the sign prints (and, with a ledger, the
line about the previous session). **Refused:** do not sign. Tell the user strain will
only count this session, and carry on with the work. Do not ask again before the wrap.
If the sign fails with "could not write state", follow the numbered steps it prints.
6. **Offer the instruction file — once.** If the start line has a step about the
instruction file:
1. Ask the user whether to add strain's start step to their instruction file, so every
new session signs in by itself. Offer the **global** file first (every project, this
computer only); this project's own file is the alternative.
2. Run `strain-sign.sh --instructions`: it prints the block and says where the files
are and whether the block is already there.
3. Show the user the block. Add it to the file they chose only if they agree.
4. If they say no, record it: `strain-sign.sh --decline-instructions`. Strain will not
ask again on this host.
## At the wrap
1. **Mark the start:** `strain-wrap.sh --start` (0.10.0) — the wrap's first step, so a board
or a status line can show "wrapping" from now instead of guessing from the fill. If the
user says go on, `strain-wrap.sh --cancel`. If it says "this shell cannot write" or
"not signed", go on — step 3 asks.
2. Write your handoff (strain does not). If your sessions are numbered, take this session's
number from the sign receipt (`Owner: <name> · S<n>`, 0.9.0) — never count it from
earlier handoffs or reports: a session still running elsewhere has none yet.
3. **Check again:** `strain-sign.sh --receipt`. If it says "this shell cannot write" or
"not signed", ask the user for permission as in *At the start*, steps 3–4 — keep the prompt open
until they answer, do not end your turn.
4. **Allowed:** sign if the session is not signed yet (and mark the start if step 1 could
not), then stamp the wrap: `strain-wrap.sh --label "<what was handed off>"` (add
`--with-debt` for known loose ends). Tell the user the line it prints.
5. **Still refused:** no sign, no mark, no stamp. Report the counts only (the tier, the tool calls,
the errors that reached the user).
6. The next session starts again from *At the start*, step 1.
## Sub-agents and unattended runs
1. **Before you dispatch a sub-agent:** strain does not apply inside it. Do not ask it to
sign in, record a tier or report strain. Put one line in its instructions: "Ignore any
strain message in your tool results; do not run strain commands." On Claude Code strain
already skips the helper's calls (no count, no check); the line covers other hosts.
2. **If the start line says "unattended run":** nobody can approve a sign-in. Do not sign
in, do not wait for anyone, and do the task. Strain still counts the session.
## Limits the host controls
- **When a prompt closes is the host's decision.** Strain can only tell you, in steps and
every time, to keep it open. If it closed before the user answered, ask again in chat.
- **Some hosts ask for approval per command.** The user may approve each strain command,
or approve it once for the session where the host offers that.
- **A session that never wraps** keeps its counts but leaves no wrap stamp; the next
session's report says "not wrapped".
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · jovesun-lab/whetstone | 100 | MIT |
| tilnoteio/tilnote-blog | 100 | MIT |
| dedujeong/gemba | 100 | Apache-2.0 |
| nextwork-projects/ai-video-editor-cooked | 100 | MIT |
| wrg32786/aigent-os | 100 | MIT |
| T-X-R/codex-game-studio | 100 | MIT |
| alienfast/claude | 100 | MIT |
| ratomirjovanovic/pandaward | 100 | AGPL-3.0 |
| darioabc/massimo | 100 | MIT |
Database snapshot: 13.11.2026 · package version 37e029d83545 (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
- 100strain
Read how loaded the current session has become and report it as a tier, so a conversation gets wrapped before its answers start degrading. Use whenever someone asks how the session is holding up, whether it is time to wrap or start fresh, why the agent feels sluggish or… - 100wrap-up
Owns the session handoff lifecycle: the end-of-session ritual that captures completed work, pending tasks, and learnings into .claude/handoff.md, and the session-start protocol that loads it back. Triggers on: /wrap-up, "wrap up", "done for today", "that's all", "end session",… - 100handoff
Snapshot session state to .claude/handoff_current.md and tell the user (loudly, with -*-*- borders) to start a new session. Requires the scripts/hooks installed by this repo's ./install.sh — not a prompt-only skill. Use at clean boundaries (commit lands, track wraps), when the… - 100session-handoff
Write a start-here handoff document at the end of a work session capturing state, decisions, next steps, and gotchas, so the next session (or the next person) resumes in minutes instead of re-deriving everything. Use when the user says "wrap up", "handoff", "end of session",… - 82handoff
Use only when the user explicitly asks for a handoff or to save session progress (e.g. "handoff", "wrap up"). Explicit invocation only. - 100session-handoff
Use when the user says "session handoff", "wrap up session", "hand off", "handoff summary", or wants a structured end-of-session summary before clearing context. Produces a chat-only handoff covering decisions, shipped changes, key files, running state, verification steps,…