AICMO

This system is **fully self-contained in GitHub** - no local servers, no external schedulers. ┌───────────────────────────────────────────────────────────────────────────────────┐ │ AUTONOMOUS LOOP │ │ │ │ ┌───────────────────────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │ │ TRIGGER │───>│ AGENT │───>│ PR │───>│ AGENT │ │ │ │SCHEDULE 8AM | State Change│ │ works │ │ created │ │ reviews │ │ │ └───────────────────────────┘ └──────────┘ └──────────┘ └──────────┘ │ │ ^ │ │ │ │ v │ │ │ state change ┌──────────┐ │ │ └─────────────────────────────────────────────────│ MERGE │ │ │ triggers next work │ (auto) │ │ │ └──────────┘ │ └───────────────────────────────────────────────────────────────────────────────────┘ ## How It Works [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#how-it-works) 8 AM UTC ──> Work ──> PR ──> Review ──> Merge ───┐ │ ┌──────────────────────────────────┘ │ v Limit reached? ──No──> Work ──> PR ──> Review ──> Merge ───┐ │ │ Yes ┌─────────────────────────────┘ │ │ v v Wait for (repeat until limit) next 8 AM 1. **8 AM cron** starts the daily N\*PRs working cycle (limiting pace to preserve tokens) 2. **Agent works** - think, research, code, deliver 3. **Agent creates PR** with changes (updates `agent/state/`) 4. **PR triggers** self-review workflow 5. **Auto-merge** when approved 6. **State change triggers** next work session immediately 7. **Repeat** until MAX\_PRS\_PER\_DAY limit reached 8. **Wait** for next 8 AM to start again **File Structure** / ├── .github/workflows/ │ ├── agent-work.yml # Work session (8 AM cron + manual dispatch) │ ├── agent-review.yml # Self-review on PR creation │ ├── agent-work-trigger.yml # Chains sessions after PR merge (requires AGENT\_PAT) │ └── process-outputs.yml # Posts outputs via integrations │ ├── agent/ │ ├── config.md # Safety limits and settings │ │ │ ├── state/ # Created by agent │ │ └── current.md # Working memory │ │ │ ├── memory/ │ │ ├── hypotheses/ # Ideas being tested │ │ ├── learnings/ # Validated knowledge │ │ └── research/ # Gathered intelligence │ │ │ ├── output/ │ │ ├── plans/ # Strategic plans │ │ └── reports/ # Analysis summaries │ │ │ ├── outputs/ # Content to post externally │ │ └── x/ # Tweet files → process-outputs.yml │ │ └── posted/ # Archived after posting │ │ │ └── integrations/ # External platform scripts │ └── x/ │ ├── post.sh # OAuth 1.0a/2.0 posting │ └── verify.sh # Verify credentials │ ├── CLAUDE.md # Agent operating instructions ├── GOALS.md # Goal definition and X API setup └── README.md ## Requirements [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#requirements) ### Claude Code [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#claude-code) * `CLAUDE_CODE_OAUTH_TOKEN` or `ANTHROPIC_API_KEY` secret * Uses Claude Personal OAuth token (`claude setup-token`) or fallback to Anthropic API Key * Details: [https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md](https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md) ### Integrations (optional) [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#integrations-optional) See `.claude/skills/publishing/SKILL.md` for platform credentials and publishing rules. ## How the Agent Learns [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#how-the-agent-learns) Each session the agent reviews what happened, adjusts based on learnings, plans next steps, and executes. ## Safety Boundaries [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#safety-boundaries) * Max PRs per day: configurable (default: 2) * No changes outside `/agent` directory * All changes go through PR review * Previous versions preserved in git history * Additional limits defined in `agent/config.md` ## Getting Started [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#getting-started) ### Prerequisites [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#prerequisites) * GitHub repository with Actions enabled * `ANTHROPIC_API_KEY` secret configured * Security: * GitHub Actions allowed to create PRs (see below) * Repository ruleset configured (see below) **Repository Ruleset** A ruleset on the `main` branch is required for security — it prevents the agent (or a compromised workflow) from pushing directly to `main`, ensuring all changes go through PRs with a visible audit trail. Go to **Settings > Rules > Rulesets > New ruleset** (or edit existing): | Setting | Value | Why | | --------------- | --------------------- | ---------------------------------------------- | | Name | main | | | Enforcement | Active | | | Target branches | Default branch (main) | | | Bypass actors | Write role | Allows repo owner to push directly when needed | **Rules to enable:** | Rule | Purpose | | -------------------- | ---------------------------------------- | | Restrict deletions | Prevent branch deletion | | Block force pushes | Protect git history | | Require pull request | All changes go through PRs (audit trail) | **Pull request rule settings:** | Setting | Value | Why | | ------------------------------------ | ----- | ----------------------------------- | | Required approvals | 0 | Allows auto-merge for bot/agent PRs | | Dismiss stale reviews on push | Off | | | Require review from code owners | Off | | | Require approval of most recent push | Off | | | Required review thread resolution | Off | | > **Note:** Setting required approvals to `0` means a PR is required (audit trail) but no human approval is needed. This enables the autonomous loop while keeping all changes visible in PR history. The agent self-reviews its own PRs via `agent-review.yml`. **Workflow Permissions** GitHub Actions must be allowed to create pull requests (required for `process-outputs.yml` to archive posted content). Go to **Settings > Actions > General > Workflow permissions**: * **Allow GitHub Actions to create and approve pull requests** — must be enabled **Full Autonomous Loop (AGENT\_PAT)** **Important:** Without `AGENT_PAT`, the autonomous loop will not continue after an agent-created PR is merged. GitHub security prevents `GITHUB_TOKEN` merges from triggering subsequent workflows. This means: * Agent creates PR → Review workflow runs → PR merges * No further workflows trigger (loop stops) To enable full automation, create a Personal Access Token (PAT): 1. Go to [GitHub Token Settings](https://github.com/settings/tokens?type=beta) 2. Generate new token (Fine-grained) 3. Select this repository 4. Grant permissions: * **Contents**: Read and write * **Pull requests**: Read and write * **Metadata**: Read-only (auto-selected) 5. Add to repository: `gh secret set AGENT_PAT` With `AGENT_PAT` configured: * Agent creates PR → Review workflow runs → PR merges with PAT * Merge triggers `agent-trigger.yml` → Next work session starts ### Manual Trigger [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#manual-trigger) To start the agent manually: gh workflow run agent-work.yml ### Monitor Progress [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#monitor-progress) * Check `agent/state/current.md` for current status (created after first run) * Review PRs prefixed with `[Agent]` * Track milestones in `GOALS.md` **Configuration** ### Configurable Settings [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#configurable-settings) Edit `agent/config.md` to modify these settings: | Setting | Default | Description | | ------------------ | ------- | --------------------------------- | | MAX\_PRS\_PER\_DAY | 2 | Maximum PR cycles allowed per day | The workflow automatically reads these values from the boundaries file at runtime. ### Modify Agent Behavior [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#modify-agent-behavior) Edit `CLAUDE.md` to change how the agent operates. ### Adjust Limits [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#adjust-limits) Edit `agent/config.md` to modify safety constraints. ### Change Schedule [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#change-schedule) Edit `.github/workflows/agent-work.yml` cron expressions. ## License [](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#license) MIT ## About Self-Managing Agent to Autonomously and Iteratively accomplish given goal/task/outcome (area independent) without human participation (Experiment 2026-01) ### Resources [ Readme](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#readme-ov-file) ### License [ MIT license](https://github.com/AICMO/Autonomous-Agent-X-Bluesky#MIT-1-ov-file) [ Activity](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/activity) [ Custom properties](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/custom-properties) ### Stars [ **10** stars](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/stargazers) ### Watchers [ **1** watching](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/watchers) ### Forks [ **2** forks](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/forks) [Report repository](https://github.com/contact/report-content?content_url=https%3A%2F%2Fgithub.com%2FAICMO%2FAutonomous-Agent-X-Bluesky\&report=AICMO+%28user%29) ## [Releases](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/releases) No releases published ## [Packages](https://github.com/orgs/AICMO/packages?repo_name=Autonomous-Agent-X-Bluesky) No packages published ## [Contributors4](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/graphs/contributors) * [](https://github.com/apps/github-actions)[**github-actions\[bot\]**](https://github.com/apps/github-actions) * [](https://github.com/apps/claude)[**claude\[bot\]**](https://github.com/apps/claude) * [](https://github.com/evios)[**evios** eiosifov](https://github.com/evios) * [](https://github.com/claude)[**claude** Claude](https://github.com/claude) ## Languages * [Python100.0%](https://github.com/AICMO/Autonomous-Agent-X-Bluesky/search?l=python) ## Footer [](https://github.com/)© 2026 GitHub, Inc. ### Footer navigation * [Terms](https://docs.github.com/site-policy/github-terms/github-terms-of-service) * [Privacy](https://docs.github.com/site-policy/privacy-policies/github-privacy-statement) * [Security](https://github.com/security) * [Status](https://www.githubstatus.com/) * [Community](https://github.community/) * [Docs](https://docs.github.com/) * [Contact](https://support.github.com/?tags=dotcom-footer) * Manage cookies * Do not share my personal information