session-source-fork — Claude Code skill: fork a session
A skill for authoring rig specs: it starts a new managed seat from a prior Claude or Claude conversation without claiming the old seat continued.
npx skills add mvschwarz/openrig --skill session-source-fork -a claude-codeWhat session-source-fork does
session_source is a member-level OpenRig field. It declares where a newly launched managed seat gets its initial conversation continuity from. v1 ships one mode, fork: a new seat starts from a prior native runtime source and is not treated as the original seat carried on.
The schema is runtime-neutral, the implementation is not. Claude forks with claude --resume <parent-id> --fork-session, Claude with codex fork <parent-id>. The new seat persists a NEW continuity token, and the parent token is never written onto it. Terminal runtime rejects session_source outright.
In v1 ref.kind accepts only native_id, and value is required for it. The schema rejects artifact_path, name, last and artifact_set before launch, so there is no artifact fallback. When the source id cannot be resolved, the error names the runtime and the missing id — a silent fresh launch is not an option. Worth checking before you write the spec: fork creates a seat, restore continues an existing one, and the two are mutually exclusive.
What it helps with
- start a new managed seat from a prior Claude or Claude session
- declare a session source in a rig spec or
rig expandpayload - pick between fork, rebuild, resume and fresh for a new seat
- combine a fork with seat handover
- surface an error when the source session id will not resolve
Example prompts for the agent
- add
session_sourceto the member withmode: forkandref.kind: native_id - run
rig expandwithsession_sourceso the new seat forks from the prior session - check the runtime is not
terminal, orsession_sourcegets rejected
How to install session-source-fork
1Skills CLI (Vercel)
npx skills add mvschwarz/openrig --skill session-source-fork -a claude-code2Manually: put the files in place
~/.claude/skills/session-source-fork/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/session-source-fork/
1Skills CLI (Vercel)
npx skills add mvschwarz/openrig --skill session-source-fork -a codex2Manually: put the files in place
.agents/skills/session-source-fork/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add mvschwarz/openrig --skill session-source-fork -a cursor2Manually: put the files in place
.cursor/skills/session-source-fork/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install mvschwarz/openrig session-source-fork2Manually: put the files in place
.github/skills/session-source-fork/SKILL.md1Skills CLI (Vercel)
npx skills add mvschwarz/openrig --skill session-source-fork -a gemini-cli2Platform CLI
gemini skills install https://github.com/mvschwarz/openrig --path skills/_canonical/core/session-source-forkManually: put the files in place
.devin/skills/session-source-fork/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add mvschwarz/openrig --skill session-source-fork -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about session-source-fork
How do I install session-source-fork in Claude?
Run npx skills add mvschwarz/openrig --skill session-source-fork -a claude-code. The skill lands at ~/.claude/skills/session-source-fork/SKILL.md.
What does session-source-fork do?
It describes how a new managed seat derives its initial conversation continuity from a prior native source. v1 covers mode: fork with ref.kind: native_id for Claude and Claude.
How is a fork different from restoring a seat?
A fork creates a new seat with a new continuity token, while restore continues an existing managed seat. The parent token is never written onto the new seat.
Source text
SKILL.md · Apache-2.0
---
name: session-source-fork
description: |
Use when authoring a rig spec member or `rig expand` payload that needs to start a new managed seat from a prior runtime conversation source — `session_source: { mode: fork, ref: { kind, value } }`. v1 supports `mode: fork` with `ref.kind: native_id` for Claude and Codex. The new seat persists a NEW post-fork token; the parent token is NEVER written onto the new seat. NOT for restoring an existing seat or for artifact-backed mental-model rebuild.
metadata:
openrig:
stage: factory-approved
sibling_skills:
- claude-compaction-restore
- session-compaction-and-restore
- agent-startup-and-context-ingestion
- agent-starters
- seat-continuity-and-handover
---
# session_source Fork
`session_source` is a **member-level OpenRig field** that declares how a
newly-launched managed seat should derive its initial conversation
continuity from a prior runtime conversation source.
v1 supports one mode: **`fork`** — start a *new* managed seat from a
prior native runtime conversation source without claiming the original
seat continued.
The schema is runtime-neutral; implementation is runtime-specific (Claude
and Codex have their own native fork commands).
## Use this when
- Authoring a rig spec member that should fork from a prior session
- Authoring a `rig expand` payload with `session_source`
- Reasoning about whether to use `fork` vs `rebuild` vs `resume` vs `fresh` for a new seat
- Composing fork + handover (seat-handover-over-fork) — see `seat-continuity-and-handover` skill
## Don't use this when
- The seat is being **restored**, not created. Restore continues an existing managed seat. Fork creates a new seat.
- The continuity is **artifact-backed mental-model rebuild** (packet-derived understanding, not native runtime continuity). Use `mode: rebuild` (see `seat-continuity-and-handover` for the rebuild surface) — do NOT collapse fork into artifact-backed reentry; the distinction is load-bearing.
- The runtime is `terminal`. Terminal runtime rejects `session_source`.
## The shape (canonical YAML)
```yaml
members:
- id: reviewer-2
runtime: claude-code # or "codex"; not valid on terminal
agent_ref: specs/agents/reviewer.yaml
profile: reviewer
cwd: .
session_source:
mode: fork
ref:
kind: native_id # v1 fork mode supports native_id only
value: "0b0165d7-cb4d-4650-90de-15c0a1ede9e6"
```
Rules:
- `mode` v1 only valid value: `fork`.
- `ref.kind` v1 supports `native_id` only. Schema rejects `artifact_path`, `name`, `last`, and `artifact_set` pre-launch with explicit deferred/weaker/wrong-mode error messages (see `packages/daemon/src/domain/rigspec-schema.ts` validateSessionSourceFork). Adapter-level refusal exists as defensive handling but should not be reachable in v1 because schema rejects first.
- `value` required for `ref.kind: native_id` in v1 (the only schema-accepted kind). Future-state value semantics for `artifact_path` / `name` / `last` are not active in v1 because schema rejects those kinds pre-launch.
- `rig expand` accepts the same shape so dynamically-added members can carry session-source attribution.
The primitive does NOT introduce a new top-level command. It flows
through existing rig spec and expansion pathways.
## State model
`session_source` is a **launch-time input**, not a long-lived stateful field:
1. **Declared** — present in member config or expansion payload
2. **Resolved** — at launch time, OpenRig resolves the `ref` against the runtime
3. **Realized** — runtime fork succeeds; new managed seat receives a **NEW** native continuity token (Claude session id or Codex thread id). OpenRig persists that NEW token. **Parent token is NEVER written onto the new seat.**
4. **Failed** — resolution or fork failed; seat not launched as a fork; clear error names the resolution step that failed
Once realized, `session_source` is essentially history. Restoring the
seat later is `restore` of the new seat, not re-fork-from-parent.
## Failure modes (5)
1. **Source session id not found** — runtime cannot resolve `native_id`. **Action**: emit error naming runtime + missing id; do NOT silently launch fresh. (Future-state note: when `artifact_path` is supported in a follow-up slice, it could become a candidate fallback for Claude; in v1 schema rejects `artifact_path` pre-launch so no fallback path exists.)
2. **Source artifact path missing** *(out of v1 scope)* — would apply once `ref.kind: artifact_path` becomes schema-accepted. v1 schema rejects this kind pre-launch.
3. **Unsupported runtime/kind combination** *(largely pre-empted in v1 by schema rejection)* — schema rejects all non-`native_id` kinds upfront. Adapter-level refusal exists as defensive handling but is not reached in v1.
4. **Fork launch failed after source resolution** — runtime command (`claude --resume <parent> --fork-session` or `codex fork <id>`) returned non-zero or hung. Preserve runtime stderr; do not record new seat as launched; do not write parent token onto seat.
5. **Persistence inconsistency** — fork succeeded but seat-token persistence cannot record new continuity token. **Internal failure**: seat is not considered launched until new token is durably written.
## Honest UX rule (verbatim)
The primitive must NOT report "restored the original agent" or "resumed
the original seat" or "snapshot." The correct framing is **"forked from
source session"** / **"started from prior conversation source."**
Check the actual user-facing outcome and persisted identity; wording alone
does not prove the intended continuity was created.
## Hard boundaries (do-not list; verbatim)
- **Do NOT introduce a DIVERGENT fork primitive.** The primitive flows through existing spec/expansion pathways. *(Update 2026-08-07: `rig fork <source-session>` has since shipped as a thin convenience verb that COMPOSES the existing agent-image fork path — this honors the boundary; it is not a divergent primitive. The rule now reads: no NEW fork mechanics outside the spec / expansion / agent-image pathways.)*
- **Do NOT report "restored" / "resumed the original seat" / "snapshot"** in any UX surface.
- **Do NOT couple `session_source` to AgentSpec.** It's a member-level launch-time input.
- **Do NOT change `restore` semantics.** `session_source` creates a seat; `restore` continues an existing managed seat.
## Adapter command shape (shipped v1)
| Runtime | Command shape |
|---|---|
| Claude (`mode: fork` + `ref.kind: native_id`) | `claude --resume <parent-id> --fork-session` |
| Codex (`mode: fork` + `ref.kind: native_id`) | `codex... fork <parent-id>` |
| Terminal | Rejected at schema level |
Mutual exclusion between `resumeToken` (restore path) and `forkSource`
(fork path) is enforced at three layers in the daemon.
## Continuity outcome literals
| Outcome | When |
|---|---|
| `forked` | Fork succeeded; new seat has new native token; parent never written onto new seat |
| `fresh` | Fresh launch (no `session_source` declared) |
| `resumed` | Restore of existing managed seat |
| `failed` | Resolution or fork failed |
For `seat handover over fork` composition, the binding outcome is
independent (see `seat-continuity-and-handover` skill).
## Verify the running implementation
A source checkout and the running daemon can contain different implementations.
Before an authorized fork, check the target daemon's build identity and support
for the requested source. A source or isolated test establishes only that cut's
behavior; it does not prove the live runtime uses it.
Record the actual parent history, new seat identity, new continuity token and
independent binding outcome. Preserve errors and incomplete results rather than
reporting a fresh launch or restored original seat as a successful fork. These
checks confer no authority to launch, replace or retire a live seat.
## Currently shipped (v1) vs deferred
Shipped at openrig `c7b6df1` (2026-04-30):
- Schema accept/reject for full Honest Refusal Matrix
- Codec roundtrip (serialize → parse → normalize preserves `session_source` faithfully)
- Expansion path through `rig expand` member-input
- Adapter command shape for Claude + Codex `native_id`
- Persistence honesty (seat's `resume_token` is the NEW post-fork token; parent token NEVER written)
- Honest UX literal contract (`continuityOutcome: forked`)
Deferred:
- Claude `artifact_path` mode (schema currently refuses with deferred message)
- Provenance columns (`parent_native_id` / `created_via` for queryable RSI consumer)
- Cross-host fork (source on host A, new seat on host B) — depends on `cross-host-rig-commands`
## See also
- `seat-continuity-and-handover` skill — sibling occupant-creation primitives (resume / fork / rebuild / fresh) + seat-binding (handover composes with fork)
- `agent-starters` skill — composes session_source fork into named reusable starting points
- `cross-host-rig-commands` skill — multi-host fork (deferred)
© mvschwarz · Apache-2.0 · mvschwarz/openrig