session-start
Skill0 starsChanged 4 months agoMITView on GitHub
Initialize a fresh work session. Reads the handoff written by session-closer, cross-checks against current project state, eagerly preloads referenced docs, detects between-session drift, and presents a scannable briefing. Use at start of a new session, or when asked "what's next", "where did we leave off", "session start", or after /clear.
42 other projects use a version of this.
Adopted by
Most recently taken up by
- maiconlara/session-starttheir own version, first seen Nov 16, 2026
- fizz/claude-skillstheir own version, first seen Nov 16, 2026
- timursafiullin/aodtheir own version, first seen Nov 15, 2026
- beworkall/agent-skills-securitytheir own version, first seen Nov 15, 2026
- AlexZio00/sovereign-skillstheir own version, first seen Nov 14, 2026
- kimkon1011/orgiast-claude-rulestheir own version, first seen Nov 13, 2026
Used with
Projects with it also use session-end, deploy-verify and 3 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
jasonmichaelbell78-creatorSkillsession-end
Also in 12 projects that use this
danielarif26Skilldeploy-verify
Verify a Vercel deployment actually went live and serves the correct production domain. Self-improving: reads its own run history before checking and logs the outcome after. Use when the user asks "did you deploy?", "is it live?", "verify the deploy", "check vercel", or after any Vercel deploy. Part of the Agentic OS (Level 1 backbone).
Also in 3 projects that use this
Microsoft AzureSkillreflect
Learning capture system that extracts HIGH/MED/LOW confidence patterns from conversations to prevent repeating mistakes. Use after user corrections ("no", "wrong"), praise ("perfect", "exactly"), or when discovering edge cases. Complements .squad/agents/{agent}/history.md and .squad/decisions.md.
Also in 4 projects that use this
ibahgatSkillautopilot
Run autonomous multi-stage execution loops until acceptance criteria pass or blockers are reached.
Also in 3 projects that use this
MicrosoftSkillhandoff
Compact the current conversation into a handoff document for another agent to pick up.
Also in 3 projects that use this
Other versions42 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 42
- bitwize-music-studio/claude-ai-music-skills★ 668same name
- AlexZio00/sovereign-skills★ 158same name
- yahor-punko/mavericks-os★ 104same name
- Kanevry/session-orchestrator★ 53same name
- xcota/pos★ 32same name
- iamneilroberts/claude-skills★ 26same name
- albertomarturelo/sii★ 19same name
- IlyaGorsky/memory-toolkit★ 13same name
- mae616/ai-template★ 9same name
- freskhu/genesis★ 5same name
32 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- atomic-plannerSkill
- freshenSkill
- session-closerSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 35 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
If a remote exists: git fetch --dry-run 2>&1 — is the local branch behind?
- Distinctive: line 40
Do NOT suggest starting a new workstream.
- Distinctive: line 42
Verify each mid-flight file still exists.
- Distinctive: line 51
If a listed file no longer exists, flag "preload file missing" and continue with the others.
- Very specific: line 121
Strictly read-only: no file writes, no commits, no git mutations.
“Common” is measured: kept by at least 6 in 10 of the 35 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Read-side of the session lifecycle. Paired with session-closer (write-side). Their shared interface is memory/NEXT_SESSION.md plus the mirrored handoff block at the bottom of the latest session log.
Strictly read-only. Writes no files. Makes no commits. Runs no git mutation commands. The first write of the new session belongs to whatever workstream the user chooses to run.
Process
Phase 1 — Gather context
- Read
memory/NEXT_SESSION.md(primary handoff). If present and parseable, extract all fields. If missing, empty, or malformed, fall back to parsing the## Next session prioritiesblock at the bottom of the latest session log (memory/sessions/YYYY-MM-DD.md, highest date). If both are missing: recordNO_HANDOFFand degrade gracefully (see Edge Cases). - Read the latest session log for narrative continuity — extract the "What was accomplished" and "Decisions made" bullets. Note the session log's date and compute age in days relative to today.
- Read
WORK_PROGRESS.mdsnapshot dashboard —Last updated:timestamp, active workstream indicator, milestone statuses, active/resolved blockers. - Read
MEMORY.mdheader —Current Status/Next:line and the most recent 2–3 Recent Sessions rows. Should agree with WORK_PROGRESS and the handoff; any disagreement is drift. Git state check (read-only, never mutates):
git status --porcelain— any uncommitted changes?git log --oneline -5— last 5 commitsgit remote -v— is a remote configured?- If a remote exists:
git fetch --dry-run 2>&1— is the local branch behind?
Never run
git pull,git fetch(without--dry-run),git merge,git rebase,git reset,git stash,git checkout <branch>. All are forbidden; session-start is strictly read-only.
Phase 2 — Verify & detect drift
- Cross-check handoff vs. current state — does the handoff's "Current state" reference a workstream matching WORK_PROGRESS's active phase indicator? If not, flag "handoff vs. WORK_PROGRESS divergence" and trust WORK_PROGRESS as source of truth.
- Handoff staleness — compute age from the handoff's
written YYYY-MM-DD HH:MMtimestamp. Flag at ≥7 days ("stale handoff — recommend /freshen before starting"). Flag more strongly at ≥30 days ("very stale — review WORK_PROGRESS manually"). - Detect between-session git drift — compare the last commit in
git logto the expected post-close commit. If there are commits ahead of the expected session-close commit, flag "unexpected commits since handoff" with a list. Ifgit statusshows dirty files not listed in the handoff's "Do not touch" section, flag "unexpected dirty state" and ask the user to reconcile before proceeding. - Checkpoint resume detection — if the handoff's "Current state" indicates checkpoint mode (workstream in progress), switch to checkpoint-resume mode:
- Do NOT suggest starting a new workstream.
- Re-read the handoff's "Do not touch" section carefully — these are the mid-flight files.
- Verify each mid-flight file still exists.
- Present a resume briefing instead of a start briefing.
- Remote behind detection — if
git fetch --dry-runshowed remote commits ahead of local, note the count in the report. Do not auto-pull. Present the fact and let the user decide — pull is a write action.
Phase 3 — Eagerly preload context
- Read every file listed in the handoff's "Context to preload" section. This is the single biggest productivity lever of session-start: the first substantive turn should already have the relevant docs loaded. Don't make the user ask "please read X first."
- Cap at 10 files. If the handoff lists more, preload the first 10 and report the rest as "not loaded — load manually if needed."
- If a listed file no longer exists, flag "preload file missing" and continue with the others.
- Also preload, unconditionally:
CLAUDE.md,WORK_PROGRESS.md(already read in Phase 1), and the workstream file for the active phase (workstreams/WS0N_*.md) if applicable.
Phase 4 — Present the briefing
- Print the session dashboard. Scannable, fits on one screen, leads with state and next action:
## Session start — YYYY-MM-DD HH:MM
### Where we are
[Current state sentence — from handoff, verified against WORK_PROGRESS]
### Last session — YYYY-MM-DD (N days ago)
- [Key finding/decision #1]
- [Key finding/decision #2]
- [Key finding/decision #3]
### Next concrete action
1. [first thing — literal command or file to open]
2. [second thing]
3. [third thing]
### Prerequisites
- [met, verified now]
- [unmet — action needed]
### Context preloaded
I've read these into the session already:
- [path] — [why it matters]
- [path] — [why it matters]
### Workstream status
[CLEAN start | CHECKPOINT RESUME from mid-flight in WSNN]
### Tree state
[clean | dirty: list of files]
[branch: main | N commits behind origin]
### Warnings
- [stale handoff (N days) → recommend /freshen]
- [unexpected commits since last session close → review with git log]
- [unexpected dirty files → reconcile before starting]
### Suggested first move
[Literal text to paste or command to run]Omit empty sections. If Warnings is empty, omit it.
Edge cases
- No handoff exists — skip Phase 2; preload only CLAUDE.md, WORK_PROGRESS.md, and the active workstream file; synthesize next action from WORK_PROGRESS's active phase. Warn: "No handoff found — consider running session-closer at the end of this session."
- First session ever — treat as no-handoff; briefing focuses on WORK_PROGRESS and CLAUDE.md/README.md; friendly tone.
- Handoff says checkpoint — switch to resume mode (step 9). Do NOT suggest
/clearor a new workstream. - Handoff references a completed workstream — flag as superseded; find the next unchecked workstream from WORK_PROGRESS and suggest that instead. The old handoff's preload list may still be relevant; preload anyway but note it's from a superseded context.
- Remote behind — report the count, suggest
git pullmanually, never auto-pull. - Uncommitted changes at start — if they match the handoff's "Do not touch" list, proceed in resume mode. If they don't, flag "unexpected dirty state" and ask the user to reconcile.
Source-of-truth precedence
When files disagree:
WORK_PROGRESS.mdwins for current phase, milestones, blockers.- The handoff wins for narrative continuity and preload hints.
git logwins for "what's actually committed."- The filesystem wins for "does this file actually exist?"
- Flag every divergence in the briefing so the user knows the state isn't perfectly consistent.
Invariants
- Strictly read-only: no file writes, no commits, no git mutations. If a fix is needed, flag the condition and suggest the appropriate skill (
/freshen,/session-closer) — never run it from here. - Eager preload is the point: skipping Phase 3 reduces session-start to a
/next-workstreamwrapper. Always preload. - Idempotency: running session-start twice in a row produces the same briefing. No state accumulates.
- Brevity: the briefing fits on one screen. Details live in the preloaded files, not in the briefing text.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

