Skip to content

mvschwarz/openrig / packages/daemon/specs/agents/shared/skills/core/agent-starters

agent-starters

Verified

Use when creating, refreshing, packaging, inspecting, promoting, or deprecating a named per-seat starting point — Agent Starter manifest authoring, the 6-state lifecycle (captured → named → inspectable → used → promoted → deprecated), provenance honesty, and refusal rules. NOT a VM image; a managed starting point composed from agent role + startup context + optional native session source + provenance.

Agent Starters

A named, reusable per-seat starting point. Composes an agent role, startup context, optional native session source, and provenance into something a user or rig can choose when creating, refreshing, or packaging a seat.

This registry-backed startup context is distinct from the rig agent-image surface and from a VM image. See the current behavior below before treating optional session provenance as an executable native session source.

Use this when

  • Creating a new seat from a known-good starting context.
  • Refreshing a seat with rig expand and starter_ref.
  • Authoring a new starter (registry entry).
  • Inspecting an existing starter's provenance, freshness, or recommended status.
  • Promoting / deprecating starters in the registry.
  • Composing a starter with a Composable Priming Pack (record manifest id/version, runtime, source session id or transcript path, ready-check evidence, freshness state).

Don't use this when

  • You want VM-style deterministic state capture. Starters don't capture VM state.
  • You want to copy provider auth material into a starter. Starters refer to session sources and context; they do NOT copy credentials.

What makes a starter valuable

A starter is valuable when it is functional — it carries the context the seat needs to do its task well. Functional is the only measure of a good starter. Size is not: a smaller starter is not a better one, and a bigger one is not worse. Whatever it took for the seat to become genuinely capable at its job is the right starter — 80K tokens or 800K.

Capture the seat as it naturally is at a functional, proven point. Don't pad it with context the seat doesn't use, and — just as important — don't strip context out to make it smaller. Size is an outcome of what the seat needed, never a target.

Never compact, summarize, or shrink a seat in order to make or "lean" a starter. There is nothing valuable in "smaller," and compaction is lossy — you would trade away the exact capability the starter exists to preserve. (Compaction is a separate last-resort step for a seat genuinely near its context limit, with its own before/after plan — never part of capturing a starter.)

Authoring lifecycle — 6 conceptual states

  1. Captured — a useful seat/session/context pattern is identified.
  2. Named — it becomes an Agent Starter with stable id and owner.
  3. Inspectable — runtime, context inputs, session source, and provenance are visible.
  4. Used — a rig member or rig expand operation starts from it.
  5. Promoted — evidence shows it is recommended for a role or bundle.
  6. Deprecated — replaced, stale, unsafe, or incompatible.

Failure modes (5)

  1. Overclaiming image semantics — UI/docs imply deterministic VM-style state capture. Say "starter," name what's included, show provenance.
  2. Hidden provenance — users can't tell what session, context, or spec a starter came from. Refuse promotion until provenance is inspectable.
  3. Stale starter — points at outdated doctrine, missing files, or invalid native session source. Inspect must report staleness honestly.
  4. Secret leakage — starter packages or displays provider auth material. Refuse. Refer to session sources and context, never copy credentials.
  5. Runtime mismatch — starter used with unsupported runtime. Refuse with a clear error.

Registry entry and current behavior

Save one entry as <registry-root>/reviewer-v1.yaml. The resolver selects an explicit root or OPENRIG_AGENT_STARTER_ROOT; absent those, it checks the home registry ~/.openrig/agent-starters and a configured fallback.

starter_id: reviewer-v1
role: Review the assigned change against its stated outcome.
context: Read the current task and the source needed to judge it.

The current resolver checks the entry shape and credential boundary, then delivers this YAML itself as one required guidance_merge startup file on a fresh start. It does not interpret arbitrary context refs, load a native conversation from this example, or seal an image. A RigSpec member combining starter_ref with session_source.mode: fork is currently rejected; use a separate supported session-source path when native continuity is the outcome.

Member usage:

members:
  - id: reviewer
    starter_ref:
      name: reviewer-v1

When a starter points at a primed session produced from a Composable Priming Pack, record:

  • manifest id/version
  • runtime
  • source session id or transcript path
  • ready-check evidence
  • freshness state

Proof matrix

SurfaceTest typeAuthority
Registry schema accepts minimal starterunitdaemon or config-layer prototype
Inspect shows provenance and included contextunit / snapshotdaemon or CLI
Member can use starter_refintegrationdaemon
Unsupported runtime or stale source refuses honestlyunit + integrationdaemon
No secret material copied into starter artifactgrep / fixturetester
Bundle can include or reference starterpackage inspectionbundle layer

Dependencies on other primitives

  • Separate path: session-source-fork — native conversation-source continuity; currently not composable with starter_ref
  • Firm: specification-system — declarative starter and member references
  • Soft: rig-bundles-and-shareable-artifacts — shareable starter packaging
  • Soft: context-engineering-and-retrieval — richer declarative context assembly
  • Soft: seat-continuity-and-handover — refresh and swap workflows over starters

Required-before-RSI

Agent Starters need queryable provenance and honest inspect output before RSI loops can rely on them for seat refresh. A workflow must be able to answer: "what starter did this seat use, what source session or context was included, and is that starter still recommended?"

See also

  • session-source-fork skill — low-level fork primitive that makes native session-based starters possible
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
discernment-nudgeanthropics/skillsAfter you give a substantive answer or draft that the user may act on — advice or recommendations, drafted artifacts such as goals, plans, pitches, proposals, or emails, estimates or projections, analysis or interpretation of data, factual claims they may rely on, or a multi-step argument — invoke this skill BEFORE finalizing your reply and then, if it applies, append 2-3 short follow-up questions, each tied to something specific in what you just produced, that help the user check key facts, probe the reasoning or assumptions, and notice missing context. Do this at most once per conversation. Skip it when the user asked a trivial how-to or simple lookup, wants a purely educational explanation, asked you only to format, convert, or assemble a file from content they provided, is writing code they will run, is doing creative writing or casual chat, or already asked you to double-check, cite, or review — the skill file explains these boundaries and the exact output format.Stars: 195kUpdated Claude CodeCodexVerifiedInstall
internal-commsanthropics/skillsA set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.).Stars: 195kUpdated Claude CodeCodexVerifiedInstall