All skills
simota avatar

/orbit

@35ffd55
by shingo imotasimota/agent-skills85 stars
15

Running autonomous loops for nexus-autoloop. Generates script sets from goals, designs operation contracts, audits live loops, and recovers state — runners that complete reliably.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/orbit

This session only. Nothing lands on disk.

referencecharter-loop-driver.md

≈1.5k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Orbit — Charter-Driven Loop Driver (nexus enact integration)

Purpose: Make Orbit Charter-native when nexus enact delegates a build-loop work package, so the loop gates on the Charter's Definition-of-Done, writes to the Charter's run-log, and resumes with it. This is a driver integration (same shape as the apex Phase 6 / summit Phase 5 drivers), not a copy of the charter / enact recipes — Orbit stays a spoke; enact stays the hub.

Read when: Orbit receives an _AGENT_CONTEXT / ## NEXUS_ROUTING delegation from nexus enact for a multi-iteration implementation package, or a goal references a Charter (docs/CHARTER.md + CHARTER.roster.yaml).

Why a driver, not a recipe

charter (repo analysis → instruction document) and enact (team construction → multi-package orchestration → ship) are hub-shaped meta-orchestration — decompose, route, aggregate. Orbit is loop-shaped — iterate one unit of work until an external gate says DONE. Cloning charter/enact into Orbit would make Orbit a second hub (violating both nexus hub-spoke and Orbit's own "never replace Nexus orchestration responsibilities") and create an enact → orbit → enact cycle, since enact already delegates build loops to Orbit. The right seam to close is narrow: teach Orbit to read the Charter slice it is handed and integrate with the Charter's DoD gate + run-log.

Input: the Charter slice (read-only)

enact passes a slice of the Charter for the one package being looped — never the whole document, never other packages.

Charter section Orbit use
§4 work package (this one) goal + atomic steps + per-package AC → Orbit operation contract
§5 roster entry owner skill + verified authorized model/effort + engine + explicit fallback_engine; resolve runtime facts via _common/CLI_COMPATIBILITY.md
§7 verification plan (package-scoped) verify.sh / DONE-gate commands
§10 per-package DoD checklist the external DONE gate (see below)
§3 conventions + §9 run-log path commit/test/lint/build commands; where to append events

These sections are sha256-pinned at loop start (Orbit AP-13 / AP-16 / AP-20 immutability rule): Orbit consumes them read-only and never edits §4/§5/§7/§10. It appends only to the §9 run-log and its own loop artifacts.

DONE gate = §10 DoD (not agent self-assessment)

Orbit already forbids agent self-assessment for termination and requires an external gate + independent critic. Under a Charter-driven loop, the external DONE gate is the §10 per-package DoD checklist:

  • Loop is DONE/SUCCESS only when every DoD item passes (AC met + tests green + lint/build/typecheck + review).
  • Partial pass → PARTIAL; unrecoverable after the recovery ladder → hand back to enact as SKIPPED(blocked, reason).
  • This composes with Orbit's existing CRITIC_MODEL gate and PLACEHOLDER_GREP / mutation checks — DoD is the contract, the critic is the independent verifier of it.

Run-log: append in enact's §9 vocabulary

Beyond Orbit's own progress.md (loop mechanics), Orbit also appends to the enact append-only run-log (default docs/CHARTER.run.log.md) for the package it owns, using enact's event names:

  • PKG_START at loop entry (pkg id, owner skill, engine)
  • PKG_RECOVER per recovery-ladder step (retry / fallback_engine / Scout+Builder / alt-owner)
  • PKG_DONE at terminal (SUCCESS | PARTIAL | SKIPPED) with the DoD result

Append-only, one line per event, written immediately, atomic (temp-then-rename). Two layers, deliberately separate: Orbit loop state (state.env / progress.md) = mechanics for Orbit's own resume; enact run-log = the package-level timeline the hub reads.

Two-level resume

  • enact resumes the run from the run-log tail (last PKG_DONE); a package already PKG_DONE is skipped.
  • Orbit resumes the loop for an in-flight package from its own checkpoint (state.env, durable execution / checkpoint-replay).
  • On enact resume, a package interrupted mid-loop is re-handed to Orbit, which replays from its last loop checkpoint — no work repeated, no re-confirmation.

Engine

Build-loop packages run on the engine assigned in §5. Verify its advertised spawn/capacity and authorized model using _common/CLI_COMPATIBILITY.md and the corresponding orchestration adapter before consuming the contract. An unreachable engine may use only the package's authorized fallback_engine; log PKG_RECOVER, preserve ACs and do not silently substitute a provider.

Boundaries (hub-spoke preserved)

  • Orbit drives one work package's loop. It does not construct the team, route other packages, or read the whole Charter — enact owns all of that.
  • No orbit → enact → orbit recursion: Orbit reports terminal status back to enact via _STEP_COMPLETE: / ## NEXUS_HANDOFF; enact owns cross-package sequencing and aggregation.
  • Orbit's standard resilience contract still holds: external termination (iteration cap / timeout / budget), idempotency keys, circuit breaker, no mid-loop mutation of pinned files.

Delegation handoff (what enact sends)

enact delegates via _AGENT_CONTEXT with:

  • Task: build-loop the §4 package <id>
  • Input: Charter slice (§4 package + §5 entry + §7 package gates + §10 DoD + §3 commands + §9 run-log path)
  • Constraints: engine + fallback_engine; file ownership (from §6); budget ceiling (from §8)
  • Expected_Output: terminal status (SUCCESS/PARTIAL/SKIPPED) + DoD result + run-log lines appended

Orbit returns _STEP_COMPLETE: with Status, Output (DoD result + artifacts), Handoff (Builder/Radar/Guardian as needed), Next (CONTINUE/VERIFY/DONE).

Source: SKILL.md on GitHub

1 alert13d5 checks · Risk SAFE
  • Gen Agent Trust Hub13d

    The orbit skill is a comprehensive framework for engineering and managing autonomous agent loops. It generates and audits shell scripts designed to execute tasks via various AI engines (Codex, Claude, Gemini). While the skill enables high-autonomy behaviors—including arbitrary command execution and bypassing tool permissions—it incorporates extensive safety controls such as budget caps, wall-clock timeouts, circuit breakers, and mandatory independent critic reviews to mitigate potential risks.

  • Socket13d

    2 alerts: gptAnomaly

  • Snyk13d

    Risk: LOW · No issues

  • Runlayer6mo

    5/11 files flagged

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at 35ffd55. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated 2 weeks ago

README badge

README badge for simota/agent-skills/orbit