Sol Foreman v0.3.0 — Trust the route, see the work

Sol Foreman v0.3.0 makes model routing and CLI worker evidence easier to trust.

What’s new

• Honest routing guidance for requested models, identity claims, inherited context, and runtime evidence, including current Claude CLI probe results.
• A conditional native-Codex wrapper for longer Claude CLI jobs where the harness supports visible work items. The wrapper is visible; the inner Claude process is not.
• Safer durable worker evidence: exclusive receipts, artifact validation, child cleanup on receipt failure, and honest process-closure reporting.
• Raw worker stdout and stderr remain byte-for-byte unchanged. Conservative URL redaction applies only to derived receipt and launcher-error metadata.
• Executable setup and cross-family release runbooks with consent gates, evidence requirements, and hygiene checks.

Known limitations

• Current Codex CLI JSONL does not expose a served-model field, so a requested model remains requested-pin evidence.
• Availability probes establish route availability at probe time, not the model serving every later turn.
• Native wrapper visibility does not prove native execution or served-model identity.

This release incorporates applicable lessons from the public fable-foreman v0.3.0 and claudemix routing study while preserving Sol Foreman systems that were already ahead.

Sol Foreman v0.2.0

Sol Foreman v0.2.0 is a quality-first hardening release for reliable delegation across native Codex subagents, model-pinned Codex CLI workers, and Claude CLI agents.

Highlights

• Adds lightweight mode for one bounded delegation and guarded program mode for long or parallel work.
• Registers original program item IDs so activity and slices cannot masquerade as completed work.
• Enforces pilot-before-fan-out, bounded verification backlog, exact progress reporting, and same-cause failure breakers.
• Requires independent slice verification and fresh assembled-candidate acceptance before whole-program completion.
• Adds portable write-scope identity, protected candidate materialization, and evidence isolation.
• Hardens CLI worker launch, process-tree closure, terminal receipts, and Windows Job Object assignment.
• Detects stale or unfamiliar model capability data without exposing account identity fields or making billable model calls.
• Adds cross-platform CI, issue forms, contribution guidance, security reporting, Dependabot configuration, and a public changelog.

Install

Ask Codex:

Use $skill-installer to install https://github.com/olsenbrands/sol-foreman/tree/v0.2.0/skills/sol-foreman globally.

Start a fresh Codex session after installation so the skill catalog refreshes.

Verification

• 65 dependency-free unit tests.
• GitHub Actions passes on Windows, macOS, and Linux with Python 3.9 and 3.14.
• Python compile check passes.
• Official skill validator passes.

See CHANGELOG.md for the detailed change list.

Sol Foreman v0.1.1

Sol Foreman v0.1.1 is a quality and verification hardening release for the
global Codex orchestration skill.

Highlights

• Derives observable verification criteria from the user's goal before every
  assignment and separates product proof from lead-owned orchestration proof.
• Routes by residual judgment and quality floor, with honest evidence labels
  for requested, self-reported, inherited, and runtime-confirmed model seats.
• Adds bounded worker tickets, disjoint write ownership, durable raw evidence,
  deliberate retry/escalation rules, and explicit process closure.
• Hardens blind Claude verification with a product-only candidate, empty MCP,
  disabled ambient customization, narrow tools, raw event streams, and
  before/after tree fingerprints.
• Makes capability discovery privacy-safe and non-billable while detecting
  missing, stale, newer, unfamiliar, malformed, or future-dated model-cache
  data.
• Adds a deterministic tree-fingerprint helper and 12 dependency-free tests.

Verification

• Three fresh Sol orchestration scenarios covered a clean build, a false
  DONE plus scope breach, and a prose-only task with defect correction and
  controller recovery.
• Three independent Opus 4.8 review passes drove additional hardening.
• A final native reviewer reproduced an overlong-model-slug crash, returned
  FAIL, and returned PASS only after the lead fixed it and added the exact
  regression case.
• 12/12 unit tests pass, both scripts compile, OpenAI's skill validator passes,
  and the final worktree is clean.
• The public v0.1.1 tag installed successfully into a clean temporary Codex
  home; the installed skill validated, compiled, and ran its non-billable probe
  from an unrelated directory.

Install

Ask Codex:

Use $skill-installer to install https://github.com/olsenbrands/sol-foreman/tree/v0.1.1/skills/sol-foreman globally.

Start a fresh Codex session after installation so the skill is discoverable.

Full comparison: v0.1.0...v0.1.1

Sol Foreman v0.1.0

Initial public release. Adds quality-first orchestration across native Codex subagents, model-pinned Codex CLI workers, and Claude CLI agents; pre-dispatch verification contracts; bounded retries; cross-family blind verification; a Fable permission gate; and a non-billable live capability probe. Install the versioned skill from skills/sol-foreman.