| name | codex-hyperagent |
| description | Run Codex with the HyperAgent Suit. Use when the user asks Codex to operate as HyperAgent, run a HyperAgent mission, produce mission telemetry, review mission friction, or propose Suit upgrades. |
| metadata | {"short-description":"Run Codex with HyperAgent mission, Workshop, and Forge loops","version":"v0.1.0-alpha"} |
codex-hyperagent
Use this skill when the user asks Codex to operate as HyperAgent, run a HyperAgent mission, produce mission telemetry, review mission friction, or propose Suit upgrades. In the HyperAgent repository itself, project instructions may require HyperAgent triage on every task.
Source Of Truth
- Product contract:
docs/hyperagent-prd.md
- Suit prompt:
hyperagent/operating-prompt.md
- Capability registry:
hyperagent/capability-registry.md
- Mission template:
templates/mission-record.md
- Upgrade proposal template:
templates/upgrade-proposal.md
- Forge review template:
templates/forge-review.md
- Upgrade backlog:
workshop/backlog.md
- Workshop rubric:
workshop/rubric.md
- Forge quality rubric:
forge/process/quality-rubric.md
- Forge audit command:
sh scripts/hyperagent.sh review forge audit
- Local helper:
scripts/hyperagent.sh
Operating Loop
- Read
hyperagent/operating-prompt.md.
- Run HyperAgent relevance triage.
- If the task is relevant, run it as a Mission. Keep scope tight and verify the result.
- Run verification through
sh scripts/hyperagent.sh check -- COMMAND when possible so evidence is recorded automatically.
- Before final response on relevant tasks, create or update a mission record with
sh scripts/hyperagent.sh mission closeout --request "..." --slug "..." --outcome "..." --risks "...".
- Run
sh scripts/hyperagent.sh mission verify --strict missions/MISSION.md before using the mission as Workshop evidence.
- If the mission exposed friction, create at least one Workshop proposal in
workshop/proposals/ from templates/upgrade-proposal.md.
- Every Workshop proposal must include an
Implementation Plan section that names the single highest-priority plan step first, then lists implementation steps, files or instructions likely to change, and verification for that first step.
- Score proposal priority with
workshop/rubric.md before recommending implementation.
- If the friction is about the quality of the Workshop process itself, create a Forge review in
forge/reviews/ from templates/forge-review.md.
- Do not mark a proposal accepted unless there is an explicit human approval decision in
workshop/decisions/.
- Report the triage decision, mission outcome, verification, record path, proposal path if any, decision path if any, and unresolved risk to the user.
Use sh scripts/hyperagent.sh sense to inspect current local context. Prefer check, sense, mission closeout, and mission verify --strict when they make artifact creation more reliable, but fill in final outcome and risk judgment yourself.
Relevance Triage
Run triage before deciding whether to write mission telemetry.
Use the full HyperAgent loop when the task:
- changes files, docs, scripts, templates, skills, evals, or product behavior,
- asks about the HyperAgent PRD, architecture, setup, install flow, skill behavior, or repo status,
- requires investigation across multiple files or commands,
- involves verification, debugging, failing checks, or repeated friction,
- could reveal an improvement to the Suit, Workshop, Forge, installer, docs, or evals,
- explicitly asks to use HyperAgent or run a HyperAgent mission.
Skip the full loop when the task is clearly isolated and low-signal:
- simple factual answer that does not depend on repo state,
- trivial one-line command,
- restating prior status without new investigation,
- small conversational clarification,
- simple wording or formatting that does not affect project behavior.
When skipping, say: HyperAgent triage: isolated one-off; no mission record written.
When in doubt inside the HyperAgent repo, run the full loop. This project is the testbed.
Workshop Review Prompt
Review the recent mission records in missions/. Identify concrete Suit friction that appears in the evidence. For the highest-value friction, create an upgrade proposal in workshop/proposals/ using templates/upgrade-proposal.md. Do not activate the upgrade. Default the activation mode to human review required.
Use workshop/rubric.md and update workshop/backlog.md when the proposal should be tracked for implementation.
Forge Review Prompt
Review recent upgrade proposals in workshop/proposals/, decisions in workshop/decisions/, accepted capabilities in hyperagent/capability-registry.md, and evals in evals/ plus scripts/verify-mvp.sh.
Run Forge reviews after proposals are accepted or rejected, evals change, release-readiness checks, or repeated missions where Workshop output looks vague, unsafe, untested, or too heavy. Judge outcome quality, proposal quality, eval quality, safety quality, and process bloat. If the process itself needs improvement, create a Forge review in forge/reviews/ using templates/forge-review.md.
Use forge/process/quality-rubric.md to score the Workshop process and identify process upgrades. When a process upgrade is warranted, generate a Workshop proposal from the Forge review with sh scripts/hyperagent.sh review workshop --forge-review PATH --title "..." --problem "...". Keep the proposal in human review required mode.
Use sh scripts/hyperagent.sh review forge audit for a concise process-health report covering proposal quality, stale or missing decisions, accepted capability traceability, and Forge audit eval coverage. The audit is read-only unless --write-proposal is passed, and generated proposals must remain human review required.
Safety Defaults
- Propose upgrades freely.
- Draft local low-risk upgrade files only when asked or when the mission requires it.
- Do not activate persistent behavior changes without human approval.
- Do not broaden filesystem, shell, network, deployment, account, or secrets access without human approval.
- Do not silently alter secrets handling.
- Default all upgrade proposals to
human review required.
- Keep persistent behavior changes
human review required.
Allowed activation modes are suggest only, draft files only, human review required, and auto-install low risk.
Accepted capabilities require a decision record in workshop/decisions/ and a traceable entry in hyperagent/capability-registry.md.
File Naming
Use lowercase timestamped slugs:
missions/YYYY-MM-DD-HHMM-brief-slug.md
workshop/proposals/YYYY-MM-DD-HHMM-brief-slug.md
forge/reviews/YYYY-MM-DD-HHMM-brief-slug.md
workshop/decisions/YYYY-MM-DD-HHMM-accepted-or-rejected-brief-slug.md