Skip to content

Your Workspace

An AIOS workspace is a git repo you clone once and work in every day. It’s structured to be:

  • Context-first — frontmatter on every file declares what it is and who can see it
  • Tier-enforced — content is default-private; you promote it deliberately
  • Agent-native — Claude Code harnesses do real operational work inside the repo

Every workspace uses the same six-folder pipeline, regardless of context (consultant or employee):

FolderPurposeDefault tierSyncs?
0-context/Charter, scope, or role/OKRsteamYes
1-inbox/Raw inputs, transcripts, brain pullsadminNever
2-work/Deliverables, working documentsteamYes
3-log/Decision log, task list, hoursadminNever
4-shared/Client-facing or company outputexternalYes
5-personal/Private scratch spaceadminNever

Content moves through the pipeline deliberately — raw capture in 1-inbox, refined work in 2-work, promoted output in 4-shared. Nothing jumps tiers automatically.


Every file in the workspace has an audience tag in its YAML frontmatter:

---
title: Sprint 1 retrospective
audience: team
---

The friendly names map to canonical sync tiers:

FriendlyCanonicalWho sees it
private, personaladminYou only — never leaves your machine
teamteamAll team members on the brain
client, companyexternalExternal stakeholders

Pure-shell validators you can run against any workspace:

Terminal window
# Run all validators
validation/validate-all.sh /path/to/your-workspace
# Individual checks
validation/check-structure.sh . # numbered spine + required files
validation/check-frontmatter.sh . # YAML frontmatter on every file
validation/check-secrets.sh . # API keys, tokens (hard fail)

Validators are wired into CI — a push that fails check-secrets.sh is blocked.


hooks/team-ops-guard.sh is a Claude Code PreToolUse hook that fires on every Write and Edit. It blocks:

  1. Files containing secrets (API keys, tokens, .env content)
  2. Admin-tier content written into team-tier directories
  3. Markdown deliverables missing frontmatter

Prevention at authoring time — before anything reaches version control.


Harnesses are Claude Code skills that live in .claude/skills/ of your scaffolded workspace. They run multi-agent workflows with adversarial verification and rubric-gated self-correction.

SkillWhat it does
/decision-auditGovernance audit of the decision log — one verifier per rule + adversarial pass
/scope-creepClassify deliverables against scope baseline, adversarial re-grade to remove false positives
/transcript-decisionsFan-out extraction from transcripts → dedup → grounding → decision-log rows
/weekly-synthesisRubric-gated weekly digest — revises until the rubric passes or budget is spent
/aios-syncReview what’s staged to push, decide what to promote, run aios push
/okf-traverseAnswer a question by traversing the local OKF bundle — works offline, no brain required

The workspace maintains a structured link graph (aios pull-bundle fetches it from the brain). Every file can carry links: in its frontmatter pointing to related decisions, tasks, or documents. /okf-traverse uses this graph to answer questions by following links rather than full-text search.


The scaffold asks once: are you a consultant or an employee?

  • Consultant: 0-context/ holds engagement charter + client scope; 4-shared/ is the client-facing surface
  • Employee: 0-context/ holds role definition + OKRs; 4-shared/ is the company-facing surface

The spine structure is identical. Only the framing and folder labels inside 0-context/ and 4-shared/ differ.