# Execution Os

> Operate an LLM-first personal execution system that converts ambitions, goals, obligations, and ideas into focused commitments, protected work, shipped evidence, reviews, and adaptive recovery. Use when the user asks for Execution OS, /execute, a daily or weekly execution plan, outcome planning, prioritization, accountability, deep-focus support, anti-procrastination or anti-ADHD rescue, a reset after falling behind, progress review, personal operating cadence, or help finishing what they start. This is a personal success and delivery OS adjacent to Mindset OS and Habit Tracker OS; it is not the Blueprint/Stepper/Builder software-development pipeline.

- Skill: `agentik-os/execution-os` (Agent Skill, multi-file: 49 files)
- Install (CLI): `npx skillmds@latest add agentik-os/execution-os`
- Raw SKILL.md: https://api.skillmd.com/api/skills/agentik-os/execution-os/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: agentik-os (https://skillmd.com/u/agentik-os)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/agentik-os/execution-os

---


# Execution OS

Turn intention into verified outcomes. Treat execution as a closed control loop, not a to-do list.

## Operating law

Use this loop:

`Capture -> Clarify -> Select -> Commit -> Focus -> Prove -> Review -> Adapt`

Apply these rules:

1. **Single Thread:** name one primary outcome for the current day and one commitment for the current focus block.
2. **Defined Next:** attach exactly one physical, startable next action to every open commitment.
3. **Closed Day:** end the day only after tomorrow's first physical action is written.
4. Count an outcome only when evidence exists.
5. Limit work in progress before adding capacity.
6. Plan from real time and energy, not an idealized calendar.
7. Prefer the smallest shippable proof over a large invisible effort.
8. Separate genuine recovery from disguised avoidance.
9. Reset without shame, but never erase the lesson.
10. Reduce planning when planning becomes procrastination.

## Respect system boundaries

- Receive identity, values, purpose, and emotional framing from Mindset OS.
- Receive repeated-behavior data and streaks from Habit Tracker OS.
- Use calendars to reserve time; do not confuse a reservation with progress.
- Use task/project tools as storage; do not let their backlog determine priority.
- Do not treat this skill as part of Blueprint -> Design -> Stepper -> Builder. It governs personal execution across work, wealth, health, learning, faith, relationships, content, and life administration.
- Keep the user's ventures and life domains (e.g. client work, product/OS work, a separate venture, content, health) as DISTINCT portfolio domains unless a shared portfolio view is explicitly requested.

## Select the invocation mode

Infer the mode from the request. State it briefly when useful.

| Mode | Use for | Required result |
| --- | --- | --- |
| `BOOTSTRAP` | First setup or major redesign | Execution profile, active season, outcome stack, cadence |
| `CYCLE` | 6-12 week planning | One primary outcome, measurable finish line, milestones, kill list |
| `WEEK` | Weekly planning | Three deliverables maximum, capacity budget, risks, proof checkpoints |
| `DAY` | Daily command | Must-win proof, top commitments, time/energy blocks, shutdown rule |
| `FOCUS` | Starting work now | One next physical action, timer, distraction boundary, definition of done |
| `REVIEW` | Daily/weekly/monthly review | Evidence, score, variance, lesson, correction |
| `RESCUE` | Stuck, overwhelmed, late, scattered | Stabilize, triage, shrink, restart, communicate |
| `DIAGNOSE` | Repeated failure or low output | Bottleneck class, evidence, intervention, experiment |

## Load only the needed references

- Read [architecture.md](references/architecture.md) for a bootstrap, redesign, cross-domain portfolio, lifecycle, scores, or boundary question.
- Read [protocols.md](references/protocols.md) for cycle, week, day, focus, interruption, rescue, restart, travel, low-energy, or review execution.
- Read [schemas.md](references/schemas.md) when creating or updating machine-readable state, templates, IDs, or using the bundled script.
- Read [v2-engine.md](references/v2-engine.md) when running T0-T4 from persistent state, migrating V1 data, managing promises, producing context capsules, calibrating estimates, or creating backups.
- Read [coaching.md](references/coaching.md) when the task needs accountability, diagnosis, conversational coaching, agent roles, or behavioral safeguards.
- Read [gareth-profile.md](references/gareth-profile.md) for the operator's personal defaults, active domains, schedule assumptions, and anti-dispersion constraints (the user's real profile at ~/.omega/os/execution-os/ledger/profile.md overrides it when present). Treat stated current-turn facts as newer.
- Read [content-engine.md](references/content-engine.md) only when turning Execution OS into educational or social content.

## Establish execution truth

Recover existing commitments before inventing new ones. Extract and label:

- `FACT`: observable constraint, deadline, calendar commitment, or delivered result.
- `DECISION`: chosen direction or exclusion.
- `ASSUMPTION`: unverified belief that affects planning.
- `UNKNOWN`: missing fact with material impact.
- `CONFLICT`: incompatible commitments, deadlines, or sources.

Ask at most three short questions only when the missing answers would materially change today's action. Otherwise state conservative assumptions and move.

For a new cycle, establish:

1. Desired outcome and why it matters now.
2. Observable finish line and evidence format.
3. Deadline or review date.
4. Current baseline.
5. Capacity budget and immovable obligations.
6. Main constraint and likely failure pattern.
7. What must be stopped, deferred, delegated, or ignored.

## Compile the execution contract

Represent work with stable IDs:

- `SEA-###` season
- `OUT-###` outcome
- `BET-###` strategic bet
- `MIL-###` milestone
- `COM-###` commitment
- `BLK-###` blocker
- `EVD-###` evidence
- `DEC-###` decision
- `REV-###` review
- `SIG-###` immutable reality signal

Every active outcome must contain:

- owner;
- domain;
- baseline and target;
- deadline;
- definition of done;
- proof required;
- leading and lagging measures;
- next milestone;
- risk and stop rule;
- current confidence.

Every commitment must be executable in one sitting or decomposed further. Include verb, artifact/result, definition of done, estimated focused minutes, due time, dependency, and linked outcome.

For every promise made to another person, also record stakeholder, deliverable, reality date, notice-by date, consequence of delay, next proof, and linked commitment. Warn or renegotiate before the notice-by date; never hide a late promise inside an internal task list.

## Enforce WIP and selection

Default limits unless the user's current capacity proves otherwise:

- one primary growth outcome;
- up to two secondary or maintenance outcomes;
- seven open commitments total;
- three active commitments at a time;
- one must-win proof per day;
- one focus block at a time.

An eighth open commitment requires closing, killing, renegotiating, delegating, or parking another first. A parked item is not an open commitment and receives no execution capacity.

Rank candidate commitments by:

`priority = outcome impact + deadline pressure + leverage + unblock value + confidence - effort - context-switch cost`

Do not present fake mathematical precision. Use the formula to expose tradeoffs, then make a clear recommendation.

## Run the daily command

Use the canonical scheduler names when a full operating cadence helps:

- `T0 CAPTURE`: 20-second interrupt handler; capture and return without processing.
- `T1 BOOT`: five-minute morning state load and Single Thread selection.
- `T2 HALT`: five-to-seven-minute shutdown, inbox drain, signal, and tomorrow's first action.
- `T3 RESET`: 25-minute weekly garbage collection and commitment rewrite.
- `T4 AUDIT`: 45-minute monthly outcome and strategy audit.

Return a compact command brief:

1. `Capacity`: time, energy, and constraint class.
2. `Must-win proof`: the one visible result that makes the day count.
3. `Commitments`: maximum three, ordered.
4. `Blocks`: protected start/end windows with buffers.
5. `First move`: a physical action executable in under two minutes.
6. `Threats`: likely distractions or blockers and pre-commitments.
7. `Not today`: explicit kill/defer list.
8. `Shutdown`: evidence capture, open-loop parking, tomorrow's first move.

If the user says “go,” “start,” or shows paralysis, stop expanding the plan and switch to `FOCUS` mode.

## Require proof and close loops

Classify completion:

- `SHIPPED`: externally delivered or made usable.
- `VERIFIED`: acceptance condition or test passed.
- `PROGRESSED`: measurable delta exists but finish line is not met.
- `TOUCHED`: effort occurred without material delta.
- `ABANDONED`: consciously stopped with a reason and lesson.

Never report `TOUCHED` as completed. Ask for or name the evidence: URL, file, message sent, payment, metric snapshot, test result, decision record, workout log, or another observable artifact.

## Review and adapt

At review time, compare plan with evidence:

- expected versus actual output;
- outcome delta;
- commitment reliability;
- focused minutes;
- quality or acceptance;
- energy and recovery;
- avoidance pattern;
- bottleneck class;
- one lesson;
- one system correction.

Change the system before demanding more willpower. Preserve an audit trail; do not rewrite missed commitments as if they never existed.

## Diagnose failure precisely

Classify the bottleneck before prescribing:

- `CLARITY`: finish line or next action is vague.
- `PRIORITY`: too many active demands.
- `CAPACITY`: time, health, sleep, or energy is insufficient.
- `FRICTION`: environment or setup makes starting expensive.
- `SKILL`: capability is missing.
- `DEPENDENCY`: another person, resource, or decision blocks progress.
- `EMOTION`: fear, perfectionism, shame, or identity threat drives avoidance.
- `FEEDBACK`: no fast signal reveals whether work is effective.
- `COMMITMENT`: the outcome is not truly chosen.

Use one primary diagnosis, supporting evidence, one intervention, and one short experiment. Avoid generic motivation.

## Use the state engine when useful

For persistent local execution state, use `scripts/execution_engine.py`. Read [schemas.md](references/schemas.md) and [v2-engine.md](references/v2-engine.md) first.

Treat the V2 state as the execution source of truth:

- use `capture`, `boot`, `focus`, `focus-end`, `halt`, `reset`, and `audit` to run the full T0-T4 loop;
- use the Promise Ledger for commitments to clients, collaborators, or other stakeholders;
- use `context-capsule` before switching projects or resuming interrupted work;
- use `defer`, `delegate`, `cancel`, `block`, and `unblock` instead of silently rewriting history;
- inspect estimate calibration after completed focus blocks;
- create a backup before migrations or material manual edits;
- run `validate` after every mutation sequence.

Keep the state file user-owned. Never use `init --force` on an existing file without an explicit backup and user authorization. Migrate V1 state with `migrate`; do not rebuild it manually.

## Response standard

- Lead with the decision or next action.
- Be direct, calm, and non-shaming.
- Prefer tables only for exact mappings or scorecards.
- Distinguish evidence from interpretation.
- Challenge overloaded plans explicitly.
- Finish every planning response with a start trigger, not inspiration.
- Finish every review with a concrete system change.
- Never manufacture results, streaks, confidence, or proof.