From requirement intake to implementation, review, documentation, and deployment — every phase as a versioned skill, running inside your own agent runtime.
One package carries a complete, AI-assisted delivery process: every phase of a user story's life runs as a versioned, reviewable skill — as a slash command directly in your agent runtime (/help lists the full catalog).
Meeting transcripts become structured Jira stories with acceptance criteria.
Codebase analysis produces reviewable implementation notes.
AI-generated first drafts with tests — in the customer's own tech stack.
Rubric-scored reviews, PR tooling, and deterministic test data.
Environment promotion, release notes, and documentation.
From drafting stories out of meeting transcripts and implementing whole epics in parallel to release pull requests, four kinds of documentation, and a local business chat about your own solution — every workflow step is a slash command.
Skills, adapters, and a Node installer — no CloudRise server in the loop, no telemetry, no phone-home. Your code, tickets, and transcripts never leave your environment.
Semantic versioning with VERSION, CHANGELOG, and immutable releases. Every skill execution leaves a schema-validated log — auditable in your own repository.
No skill contains a customer name, a vendor tool, or a runtime primitive. Five independent axes are resolved from configuration when a skill executes — a new platform, runtime, or git strategy is a new adapter document, never a fork of the skills.
Identity, credentials, naming conventions, locale, and paths — your own configuration repository, linked at install time.
Stack-specific commands and standards — Salesforce, Node/Cloudflare, and more. Platform-specific skills abort cleanly where they do not apply.
Claude Code, OpenAI Codex, Gemini CLI, local LLM, or a hosted endpoint — switchable per run, with cross-runtime review.
Cloud (MCP tools) or Data Center (REST with a token) — the same Jira and Confluence workflows in both worlds.
Feature-branch, trunk-based, or gitflow — branching, PR gates, and story promotion follow your model.
A skill carries no customer facts. On every prompt it resolves the active customer and reads that customer's configuration set — and the platform's best practices — before it does anything else. Same skill, different customer: only the links point elsewhere.
Customer conventions override platform best practices on style and naming. On correctness, security, resource limits, and deployability the platform wins — a convention cannot waive that. A customer document that appears to is treated as a documentation bug and raised, never silently obeyed.
Customers ship their own skills in their own configuration repository; a skill folder named like a shipped skill replaces it — for that customer only. setup.sh <customer> switches the whole context in seconds.
Every configuration read goes through a single tool — no skill parses Markdown on its own. A new runtime, platform, or git strategy is a new adapter folder, never a change to the skills.
The skills exist exactly once — and run natively where you work. Data-sovereign too: with a local LLM or a European endpoint of your choice as the data perimeter.
Full capability: parallel subagents, worktree isolation, MCP, web search.
Native skill execution with reduced parallelism.
Native MCP, sequential processing.
Ollama / llama.cpp — for privacy and offline scenarios.
Any OpenAI-compatible endpoint — e.g. Scaleway, OVHcloud, IONOS, STACKIT. The API key never enters configuration, only the name of its environment variable.
The score is computed, never summed in prose: a rubric scorer reads the rubric and the round's findings file and emits per-axis subscores — an unmeasured axis is null, not a pass. The gate is deterministic: it decides pass, continue, or stop from counted blockers and majors, the score threshold, and the round cap.
Every skill announces its scope before doing anything else. Platform and runtime guards abort cleanly instead of guessing. A story gate refuses design, implementation, and close-out while questions are open or acceptance criteria unverified. Secrets never enter configuration.
Each skill execution writes a schema-validated JSON log into your own repository — with a verdict per check (pass / fail / unchecked), the review loop (rounds, per-axis score, exit reason), and the knowledge dose. /pipeline-stats reads them, /improve-skills acts on them.
Every run leaves a log, every review leaves findings. /improve-skills mines that evidence, clusters what recurs, and proposes the exact edit to the document that should have prevented it.
A fix to a skill or a platform rule reaches every customer with the next release; a convention or domain rule stays with the customer it came from. Every proposal names which of the two applies.
Logs and reports live in each customer's own repository. Every quote is verified against its source before a human sees the proposal — a proposal without verifiable evidence is withdrawn, not softened.
Approval is per proposal, never all-or-nothing. Anything that would weaken a check is flagged and argued separately. A loop that exhausts its budget is a reason to fix the skill — not to raise the cap.
/implement-us now demands a named test per new branch and checks diff hygiene before every commit; /design-us keeps confirmed assumptions out of the open-questions gate.A local business chat about your own solution — grounded, cited, honest about gaps. And a loop that maintains the knowledge base with every story.
/ask answers one question, /serve-chat serves the same engine as a localhost chat UI — no cloud service in between. Answers come only from the built index; every citation links to its source, and a second runtime verifies each answer against the sources.
The same engine derives a role-tailored curriculum for new team members — business case and project history first, mechanics after — with briefings and quizzes, progress included.
Unanswerable questions land in a gap log for the next /build-knowledge run. Every changed component is mapped to the documents it affects — during design and after implementation — and tracked as knowledge debt until cleared. Each run logs its knowledge dose next to its quality score.
The harness and your configuration repository are both private. Onboarding sets up two access grants — verify both once per machine before installing.
cloudrise-license.json — delivered together with your invoice@aethon-x/harness is a private npm package. Your npm account is invited to the aethon-x organization; after that, a single npm login per machine is enough. Important: without access, npm reports 404 Not Found — that is npm hiding a private package, not a wrong package name. If you are missing the invite, contact CloudRise.
On first run, harness init clones your configuration repository — from GitHub by default. If it lives elsewhere (your own GitHub organization, GitLab, Bitbucket, an internal git server), point the environment variable PIPELINE_CUSTOMER_REPO_URL at any git URL for the first init. Afterwards the repository updates via a plain git pull.
npm install --save-dev @aethon-x/harnessPut the cloudrise-license.json you received into the project root (or into ~/.config/cloudrise/). Validation runs offline via an Ed25519 signature — nothing is transmitted.
harness init clones and links your configuration, selects runtime and git strategy, and makes the skills available as slash commands. The command is idempotent — simply re-run it after every upgrade or configuration change.
npx harness init <your-config-name>npx harness doctorThen start your agent runtime from the project root — the skills appear as slash commands (/help lists them).
Four commands confirm everything is in place — each one names the next step if it fails.
npx harness version → AI Harness 2.2.0 (@aethon-x/[email protected]) npx harness license → License VALID — licensee, seats, and major version are shown npx harness doctor → "All checks passed." — every failed line names the command that fixes it npx harness init --dry-run → Preview of what a re-init would change — nothing is written
Minor and patch releases are included in your major version. Updating is a two-step:
npx harness upgrade # latest release within your licensed major npx harness init <config> # re-materialize the working structure afterwards
The Aethon-X Harness is a library of AI-assisted delivery workflows — from requirement intake to implementation, review, documentation, and deployment. Every phase is packaged as a versioned skill and runs as a slash command inside your own agent runtime.
No. The harness ships as text — skills, adapters, and a Node installer — and runs entirely inside the agent runtime you already use. There is no CloudRise server in the loop, no telemetry, and no phone-home. Even the license check runs offline.
Claude Code (recommended), OpenAI Codex CLI, Google Gemini CLI, local LLMs (Ollama / llama.cpp), and any hosted endpoint that speaks the OpenAI Chat Completions API — such as Scaleway, OVHcloud, IONOS, or STACKIT. Skills exist once and adapt at run time.
Yes. By default the configuration repository is cloned from GitHub, but any git URL works — your own GitHub organization, GitLab, Bitbucket, or an internal git server. The URL is only needed for the initial clone; afterwards the repository updates via a plain git pull.
The license is perpetual per major version; minor and patch updates are included. The license file carries an Ed25519 signature that the CLI verifies offline against a public key compiled into the package — nothing is transmitted, and no license server is contacted.
Through its own logs. /improve-skills mines skill execution logs and review reports, clusters recurring failure modes from three independent occurrences upward, attributes each to the document that should have prevented it — skill, platform best practices, customer conventions, or domain knowledge — and proposes the exact text change. Every edit is approved individually; skill and platform fixes reach every customer with the next release, customer conventions stay with the customer.
Find out in a no-obligation conversation how the harness accelerates your delivery process — including a live demo in your team's workflow.
Book a free consultation