three-shifts
Skill112 starsChanged 3 months agoMITView on GitHub
Daily standup engine. Plans tasks 3x daily (6 AM, 2 PM, 10 PM) and delivers them for approval. Execution happens in the main session via direct conversation. Night shifts auto-approve carryover from earlier in the day.
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- cron-packsSkill
- memory-upgradeSkill
- mission-controlSkill
- everclawSkill
- agent-chatSkill
- bagmanSkill
- night-shiftSkill
- pii-guardSkill
- prompt-guardSkill
- relationshipsSkill
- skillguardSkill
- xmtp-comms-guardSkill
- three-shiftsSkill
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
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
- could be automatic: a setting could enforce it
- Distinctive: line 25
No file-based state machine.
- Distinctive: line 91
All times in user's local timezone.
- Very specific: line 131
If shifts/tasks.md has items from a previous shift that weren't completed:
- Distinctive: line 150
When the user approves a shift plan:
- Distinctive: line 153
If interrupted by higher-priority work, remaining tasks carry over to the next shift
- Distinctive: line 175
No external communications (Signal, email, social)
- Distinctive: line 177
No destructive operations (rm, force push, branch delete)
- Distinctive: line 178
No security changes (key rotation, permissions)
- Very specific: line 193
All use venice/kimi-k2-5, isolated sessions, deliver via Signal.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Three Shifts v3 — Daily Standup Engine
Architecture: Plan 3x daily → deliver for approval → execute in main session.
Why v3: V2 had a 15-minute autonomous cycle executor that never worked — state machine deadlocked for 19 days, burning ~27M tokens on no-ops. V3 removes it entirely. Plans are generated by cron, execution happens through normal conversation when the user approves.
How It Works
- Planner cron fires (6 AM, 2 PM, 10 PM) — reads context, generates a prioritized task list
- User reviews — approves, modifies, or skips
- Agent executes in main session — when the user says "go" or gives specific instructions
- Daily memory captures what was done — next planner reads it for continuity
No file-based state machine. No autonomous executor. The main session IS the executor.
File Structure
shifts/
├── tasks.md # Current shift plan (generated by planner)
├── context.md # Stack, constraints, known pitfalls
└── state.json # Shift metadata (which shift, approval status)tasks.md
Generated by the planner each shift. This is a task list, not a state machine.
# Morning Shift — 2026-03-13
Approved by: user | Approved at: 6:15 AM CST
## P1 (Must do)
1. Fix bootstrap key server (keys.everclaw.xyz returning 404)
2. Test installer on fresh Linux environment
## P2 (Should do)
3. Update MEMORY.md with yesterday's work
4. Review open PRs across EverClaw repos
## P3 (Could do)
5. Memory file maintenance (archive >7 day old dailies)
## Blocked/Waiting
- XMTP agent-to-agent — blocked on BASE_WALLET_PRIVATE_KEYstate.json
Minimal metadata for the planner to read between shifts:
{
"shift": "morning",
"date": "2026-03-13",
"status": "approved",
"approvedAt": "2026-03-13T06:15:00-05:00",
"approvedBy": "user"
}Status values: planned | approved | skipped
context.md
Living document of constraints and lessons. Updated by the agent when something new is learned.
Shifts
| Shift | Cron | Window | Character |
| Morning ☀️ | 6:00 AM | 6 AM – 2 PM | Comms, decisions, urgent items |
| Afternoon 🌤️ | 2:00 PM | 2 PM – 10 PM | Deep work, building, coding |
| Night 🌙 | 10:00 PM | 10 PM – 6 AM | Research, maintenance, cleanup |
All times in user's local timezone.
Planning Phase
Model: venice/kimi-k2-5 (cost-efficient, good at structured plans)
See PLANNER.md for the full planner instructions.
What the planner reads:
shifts/tasks.md— previous shift's plan (check for incomplete items)shifts/context.md— rules, constraints, pitfallsshifts/state.json— previous shift metadatamemory/daily/YYYY-MM-DD.md— recent activityMEMORY.md— active projects (⚡ Active Context section)
Plan format:
☀️/🌤️/🌙 [SHIFT] SHIFT PLAN
Date: [DATE] | Window: [START]–[END] CST
P1 (Must do):
1. [Task] — [Est. time] — [Why now]
P2 (Should do):
3. [Task] — [Est. time]
P3 (Could do):
5. [Task] — [Est. time]
Blocked/Waiting:
- [Item] — [Blocker]
Reply: "Approve" / "Skip" / "Add: [task]"Carryover
If shifts/tasks.md has items from a previous shift that weren't completed:
- Carry forward into the new plan, marked as
[carryover] - Night shift: auto-approve carryover from morning/afternoon (no user ping)
Approval flow
| User says | What happens |
| "Approve" | Plan written to tasks.md, state.json updated |
| "Add: [task]" | Task added to plan, then approved |
| "Skip" | State.json set to "skipped", next planner picks up |
| No response (30 min) | Plan stays, next shift carries forward |
Execution
Execution happens in the main session through normal conversation. There is no autonomous executor.
When the user approves a shift plan:
- The agent works through the tasks in priority order
- Progress is captured in
memory/daily/YYYY-MM-DD.md - If interrupted by higher-priority work, remaining tasks carry over to the next shift
The planner's job is to surface the right priorities — not to control execution.
Safety Rules
Always OK (any shift)
- Reading files, organizing workspace
- Research and analysis
- Code review, documentation
- Memory updates, planning
- Git status, non-destructive commands
Requires Approval (via shift plan)
- Sending emails, messages, social posts
- Creating PRs, pushing code
- Financial transactions
- Any external-facing action
Night Shift Constraints
- No external communications (Signal, email, social)
- No financial transactions
- No destructive operations (rm, force push, branch delete)
- No security changes (key rotation, permissions)
- P3 maintenance tasks (memory, cleanup, docs) are OK autonomous
Cron Setup
Three planner jobs. No executor job.
Morning ☀️: cron 0 6 * * * (America/Chicago)
Afternoon 🌤️: cron 0 14 * * * (America/Chicago)
Night 🌙: cron 0 22 * * * (America/Chicago)All use venice/kimi-k2-5, isolated sessions, deliver via Signal.
See references/config.md for customizing schedule, models, and weekend behavior.
Version History
- v3.0.0 (2026-03-13) — Removed broken cycle executor. Simplified to daily standup engine. Plans generated by cron, execution in main session.
- v2.0.0 (2026-02-27) — Added 15-minute cycle executor (state machine). Never worked — state deadlock caused 1,527 no-op runs over 19 days.
- v1.0.0 (2026-02-22) — Single long session per shift. Timeouts killed entire shifts.
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.

