remora v0.1.16

What's Changed

• docs: add Ko-fi sponsorship details by @Nanako0129 in #14
• chore: switch funding link to Patreon by @Nanako0129 in #15
• fix: show the Patreon icon in GitHub Sponsors by @Nanako0129 in #16
• feat(orchestration): calibrate outcome verification by @Nanako0129 in #17

Full Changelog: v0.1.15...v0.1.16

remora v0.1.15

What's Changed

• fix(plan): bound readiness by executable slices by @Nanako0129 in #13

Full Changelog: v0.1.14...v0.1.15

remora v0.1.14

Main-session fallback isolation

Isolate remora child main sessions from inherited Claude fallback routing. The child-only additional settings now carry an exact fallbackModel: [] alongside the sorted availableModels allowlist. Caller settings may include fallbackModel only as an array of non-empty strings (including the valid empty array); remora validates without echoing values, discards the caller key, and injects its authoritative empty array while preserving unrelated settings, hooks, permissions, environment, and guarded-file handling.

Reject separated, equals-form, missing, and duplicate caller --fallback-model options before token resolution, context discovery, temporary settings creation, or child execution. Prompt values and prompt-file paths that resemble the option remain valid, parsing still stops at --, and explicit --model, -m, and equals-form model selection—including Terra—remains unchanged. Regression coverage retains the guarded 0600 file, cleanup watcher, argv and dry-run secrecy, Happy prompt composition, and named-role model/effort contracts.

Behavioral compatibility testing covered Claude Code 2.1.216 and 2.1.217. Managed organization policy still has higher precedence and can enforce fallback beyond remora's control. Existing installations must upgrade to v0.1.14 before this child-only policy takes effect.

Full changelog: v0.1.13...v0.1.14

remora v0.1.13

Highlights

• Align the session-only orchestration policy with pilotfish v1.3.0 using backend-neutral recurrence conditions, known-remedy Execution handling, coherent fresh-verification boundaries, Plan convergence, and pull-based subagent result collection.
• Let Happy and similar wrappers reuse remora routing without losing caller --settings, hooks, permissions, or system prompts. REMORA_COMPOSE_SYSTEM_PROMPT=1 combines caller content with remora's policy while preserving -- and hyphen-prefixed prompt values.
• Keep remora-owned gateway, authentication, model, subagent, Fast, context, concurrency, effort, tool-search, coralline, prompt-bridge, and session environment keys authoritative.
• Move merged caller settings out of child argv and dry-run output. A guarded 0600 temporary file is removed after normal exit, launch failure, dry-run, or process termination.

Linux release hotfix

Large inline settings documents can exceed Linux filename limits during the initial file-or-JSON probe. v0.1.13 treats ENAMETOOLONG as an inline JSON candidate while preserving actionable errors for real settings files and the existing strict JSON validation.

The immutable v0.1.12 tag stopped in the release workflow before publishing a GitHub Release or assets. v0.1.13 supersedes it and contains the complete policy and Happy integration scope.

Upgrade

Use the approval-gated runbook pinned to this release:

https://raw.githubusercontent.com/Nanako0129/remora-cc/v0.1.13/install/AGENT-INSTALL.md

Verification

• 72 unit tests plus installer and bootstrap isolation checks
• Actual Claude Code caller-settings handoff
• Normal-exit and SIGKILL temporary-file cleanup
• Release archive checksum and GitHub artifact attestation
• Clean-room install with the native ~/.claude manifest unchanged
• Fresh adversarial verification: CONFIRMED
• Codex review: clean on PRs #8, #9, #10, and #11

Pull requests

• #8 — pilotfish v1.3.0 orchestration semantics
• #9 — caller settings and system-prompt composition
• #10 — v0.1.12 release preparation
• #11 — Linux inline-settings hotfix and v0.1.13

Full changelog: v0.1.11...v0.1.13

remora v0.1.11

What's Changed

• feat(runtime): add session-scoped Fast mode by @Nanako0129 in #7

Full Changelog: v0.1.10...v0.1.11

remora v0.1.10 - complete pilotfish role shadowing

Complete pilotfish role shadowing

remora v0.1.10 makes its session-scoped agent roster complete. A global pilotfish installation can no longer fill missing plan-verifier or security-reviewer roles with user-level definitions inside a remora session.

What changed

Area	Behavior
Session roster	All eight current pilotfish role names are supplied through remora's dynamic --agents document.
Plan review	plan-verifier is a tool-allowlisted read-only role and returns only READY or REVISE.
Outcome review	verifier retains command execution for independent reproduction and returns only CONFIRMED or REFUTED.
Security review	security-reviewer gathers read-only evidence before approval; security-executor accepts only approved implementation work.
Existing configs	Six-role TOML files remain valid through verifier and security-executor routing fallbacks. Explicit new keys allow independent effort tuning.
Native Claude	remora still does not modify ~/.claude; the complete roster exists only in the child session.

Internal routing fallback metadata is removed before the JSON reaches Claude Code. Regression coverage locks the eight-role roster, capability boundaries, launch payload, and upgrade behavior.

Verification

• 40 Python tests passed.
• Installer and offline bootstrap isolation tests passed.
• Python and shell syntax checks passed.
• Agent rendering and sanitized dry-run checks passed.
• The v0.1.10 release archive and checksum were built locally and its layout inspected.
• remora doctor rendered all eight definitions successfully.
• Codex reviewed implementation PR #5 and gave a clean pass.

Upgrade

Use the pinned, approval-gated v0.1.10 runbook:

Read and follow this installation runbook:
https://raw.githubusercontent.com/Nanako0129/remora-cc/v0.1.10/install/AGENT-INSTALL.md

Perform only the read-only preflight first. Show every proposed filesystem
change, trust boundary, download source, and verification step. Do not write
anything until I explicitly approve.

After upgrading, start a fresh remora session. /resume may restore the agent definitions recorded in an older transcript.

Full Changelog: v0.1.9...v0.1.10

remora v0.1.9 - coralline usage-state isolation

coralline usage-state isolation

remora v0.1.9 prevents GPT gateway usage data from leaking into coralline's native Claude rate-limit and burn segments.

A remora child uses the host's Claude Code status line, but its rate_limits values belong to the configured GPT gateway account rather than the host's native Anthropic login. When coralline cross-session sync was enabled, both environments previously shared the same high-water and burn stores. A gateway percentage could therefore pin the native 5-hour or 7-day segment, and gateway samples could distort the native burn ETA and slope.

What changed

Area	Behavior
5-hour and 7-day limits	CORALLINE_RL5H_FILE and CORALLINE_RL7D_FILE now point at gateway-scoped stores.
Burn estimator	CORALLINE_BURN_FILE is isolated with the limit stores, so gateway samples cannot affect the native burn slope or ETA.
Gateway identity	Store directories use a bounded readable host prefix plus a hash of the full gateway URL. Scheme, host, port, and path all participate, so path-routed accounts remain separate without exceeding filesystem component limits.
User config	A generated wrapper preserves the original CORALLINE_CONFIG and VL_CONF identity while loading the user's config, then reapplies only the scoped store paths. Existing options, sibling-file lookups, and deferred helpers retain native semantics.
Native Claude isolation	Runtime files live under ${XDG_STATE_HOME:-$HOME/.local/state}/remora-cc/coralline/. remora does not write these files into ~/.claude.
Uninstall	Default uninstall removes the known coralline runtime subtree while preserving config.toml, including when XDG state and config homes point at the same directory.
Diagnostics	remora dry-run shows the generated wrapper and all scoped coralline paths.

This patch does not change model routing, agent roles, gateway authentication, context-window policy, or the phase-aware orchestration lifecycle introduced in v0.1.8.

Verification

• 38 Python tests passed.
• Installer and offline bootstrap tests passed.
• A real launcher-to-fake-Claude test verified that the native ~/.claude manifest and settings checksum remain unchanged.
• Normal and aliased-XDG uninstall cleanup passed.
• Python and shell syntax checks passed.
• Agent rendering and sanitized dry-run checks passed.
• The v0.1.9 release archive and checksum were built locally.
• Codex reviewed PR #3 and gave a clean pass.

Upgrade

Use the pinned, approval-gated v0.1.9 runbook:

Read and follow this installation runbook:
https://raw.githubusercontent.com/Nanako0129/remora-cc/v0.1.9/install/AGENT-INSTALL.md

Perform only the read-only preflight first. Show every proposed filesystem
change, trust boundary, download source, and verification step. Do not write
anything until I explicitly approve.

After upgrading, start a fresh remora session. Existing remora processes keep the child environment captured when they launched.

Full Changelog: v0.1.8...v0.1.9

remora v0.1.8 — phase-aware Baton composition

Phase-aware Baton composition for GPT-5.6

remora v0.1.8 adds a complete plan-first orchestration lifecycle to the existing session-scoped GPT-5.6 role map:

Discovery → main-session Plan → explicit approval → Execution → fresh verification

The launcher remains isolated: model routing, gateway credentials, agents, and orchestration policy exist only inside the remora child process. Native Claude settings and login state remain untouched.

Included changes

Area	Behavior
Discovery	Luna Explore and scout roles may gather bounded evidence before the implementation outcome is known.
Plan	The Sol verifier can challenge a material Plan using READY or REVISE; the main session owns synthesis and revisions.
Approval	Large, architectural, risky, or explicitly plan-first work waits for explicit approval before implementation.
Execution	Luna handles mechanical volume work; Sol remains available for local engineering judgment and security-sensitive work.
Outcome verification	A fresh Sol verifier returns CONFIRMED or REFUTED after implementation.
Baton composition	Baton chooses questions, topology, ownership, sequence, budgets, and stop conditions while remora retains role and model authority.
Dispatch brake	Small repository scans and tightly coupled debugging stay in the main session; stable repetition and substantial independent evidence remain eligible for delegation.
Model routing	Named roles continue to omit invocation-level model; the session-scoped role definition is the sole model source.
Packaging	Baton compatibility evidence is included in the release archive.

This release keeps remora's existing six-role, session-scoped architecture. It does not install global agents or modify native Claude state.

Public compatibility gate

A fresh two-turn Gate ran through remora, Calico Claude, CLIProxyAPI, GPT-5.6, and the installed Baton skill:

Turn	Wall time	Client-reported cost	Result
Discovery + Plan	294.102 s	$1.233720	Two background Luna discovery agents; Sol readiness review returned REVISE, then READY; zero writes before approval.
Approved execution + verification	248.304 s	$0.915572	Baton correctly rejected a needless writer; the main session wrote only REPORT.md; tests passed; fresh Sol verifier returned CONFIRMED.
Total	542.406 s	$2.149292	Complete lifecycle passed without startup resend.

All four named Agent calls omitted invocation-level model. Exploration ran on GPT-5.6 Luna and verification ran on GPT-5.6 Sol.

The rejected first candidate is also disclosed: it used the wrong verdict vocabulary during Plan review and later exposed citation gaps. It is preserved as rejected evidence rather than counted as a successful compatibility run.

Exact prompts, normalized Agent calls, model routing, hashes, timing, startup-stall criteria, and limitations are published under benchmarks/baton-compatibility.

These are single-run observations, not population estimates or provider invoices.

Verification

• 34 Python tests passed.
• Installer isolation and offline bootstrap tests passed.
• Python and shell syntax checks passed.
• Agent rendering and sanitized dry-run checks passed.
• The v0.1.8 release archive was rebuilt by GitHub Actions.
• checksums.txt and build-provenance attestation were published.
• Codex review found no major issues.

This release does not modify CLIProxyAPI or gateway configuration.

Upgrade

Use the pinned, approval-gated v0.1.8 runbook:

Read and follow this installation runbook:
https://raw.githubusercontent.com/Nanako0129/remora-cc/v0.1.8/install/AGENT-INSTALL.md

Perform only the read-only preflight first. Show every proposed filesystem
change, trust boundary, download source, and verification step. Do not write
anything until I explicitly approve.

After upgrading, start a fresh remora session. /resume may restore an older session's previous agent definitions and orchestration policy.

Full Changelog: v0.1.7...v0.1.8

remora v0.1.7

Codex runtime-catalog hot update resilience

remora v0.1.7 is a focused context hotfix. CLIProxyAPI and the bundled Codex catalog can temporarily keep an older 372K value after the ChatGPT-authenticated Codex runtime catalog changes GPT-5.6 Sol, Terra, and Luna to 272K. Previous Calico sessions trusted the gateway value alone and could exceed the backend's actual context window.

This release keeps the two authorities visible and uses the smaller per-model value for the Calico client:

provider_window=372000
codex_window=272000
client_window=272000
effective_window=258400
auto_compact_window=272000
compact_policy=exact:244800

Included changes

• Read fresh context metadata from the local Codex models_cache.json; no Codex auth files are opened.
• Keep CLIProxyAPI's catalog as a separate gateway diagnostic ceiling.
• Fall back conservatively to 272K when the Codex cache is missing, older than its 300-second TTL, or incomplete.
• Restore a later fresh 372K runtime value automatically; 272K is not a permanent model pin.
• Cap explicit Calico compact-window overrides to the resolved client ceiling.
• Report gateway, Codex runtime, and final Calico model windows separately in remora doctor --online.
• Update the bilingual context, gateway, architecture, installer, and security documentation.

This release does not include the draft Baton-inspired orchestration work in PR #1. That work remains independently gated for v0.1.8.

Verification

make check passed 29 Python tests, including 372K → 272K capping and a later fresh 372K rollback, plus installer isolation, bootstrap/package, Python and shell syntax, and dry-run checks. make package completed and the archive contents were inspected.

A real non-persistent Calico/gateway request completed successfully on gpt-5.6-sol, returned REMORA_CONTEXT_SMOKE_OK, and reported contextWindow: 272000. The live doctor also confirmed the Codex active-turn bridge remained protocol v1 ready.

The GitHub release workflow rebuilt the archive, published checksums.txt, and attached build-provenance attestation.

Upgrade

Use the pinned, approval-gated v0.1.7 runbook:

Read and follow this installation runbook:
https://raw.githubusercontent.com/Nanako0129/remora-cc/v0.1.7/install/AGENT-INSTALL.md

Perform only the read-only preflight first. Show every proposed filesystem
change, trust boundary, download source, and verification step. Do not write
anything until I explicitly approve.

After upgrading, start a fresh remora session. A resumed session may retain its previous context and agent definitions.

Full Changelog: v0.1.6...v0.1.7

remora v0.1.6

Role-owned model routing

remora now makes the Agent invocation boundary explicit: every existing named role must be invoked without a model argument. Its session-scoped --agents definition is the sole model source.

This fixes the precedence failure where a per-call alias such as sonnet could override the configured role map—for example, turning a Luna executor into Terra despite remora agents showing Luna. Only a truly ad-hoc agent with no named role definition receives an explicit invocation model.

Included changes

• Named Explore, scout, mech-executor, executor, verifier, and security-executor calls omit invocation-level model.
• The same rule applies to any other existing named role definition; its definition remains authoritative.
• Truly ad-hoc agents still set model explicitly so they do not inherit Sol accidentally.
• The contract is child-session-only and does not modify native Claude settings, CLIProxyAPI, or Calico.
• English and Traditional Chinese model-routing documentation now explains Claude Code's invocation-over-frontmatter precedence.
• The approval-gated installer reports the model-ownership contract before installation.
• Regression tests verify both named-role ownership and the ad-hoc exception alongside the existing model allowlist, session isolation, background scheduling, context, and active-turn checks.

Verification

make package passed 24 unit tests, clean-room installer isolation, checksum and archive-safety bootstrap tests, Python and shell syntax checks, dry-run secret redaction, and release packaging. The GitHub release workflow rebuilt the archive, published checksums.txt, and attached build-provenance attestation successfully.

The maintainer installation was also updated to remora 0.1.6; the existing config.toml and native ~/.claude path manifest were preserved.

Upgrade

Use the pinned, approval-gated v0.1.6 runbook:

Read and follow this installation runbook:
https://raw.githubusercontent.com/Nanako0129/remora-cc/v0.1.6/install/AGENT-INSTALL.md

Perform only the read-only preflight first. Show every proposed filesystem
change, trust boundary, download source, and verification step. Do not write
anything until I explicitly approve.

After upgrading, start a fresh remora session so it receives the new orchestration policy. A resumed old session may retain its previously recorded session definitions and instructions.

Full Changelog: v0.1.5...v0.1.6