Repository navigation
feat(cli): freeze versioned harness modes - #55
Merged
Merged
Conversation
Give research runs canonical versioned profiles and reserve the serve-backed runtime names without claiming unmerged support. The registry is executable, immutable, and self-hashing so episode records can identify the exact treatment definition.\n\nCo-authored-by: OpenAI Codex <codex@openai.com>
Make --harness-mode the drift-resistant entry point for frozen research treatments, reject unavailable serve-backed modes honestly, and record the resolved profile plus reproducibility provenance in episode.json.\n\nCo-authored-by: OpenAI Codex <codex@openai.com>
Require pinned models, reject dirty source checkouts, record a deterministic runtime code-tree digest, and keep mutable availability outside immutable mode hashes.\n\nCo-authored-by: OpenAI Codex <codex@openai.com>
xisen-w
added a commit
that referenced
this pull request
Oct 19, 2026
* chore(cli): refresh the 0.1.16 release notes for #51-#61 The 0.1.16 notes were finalized in #49, but #51, #55, #56, #58, #59 and #61 merged after that into the same unpublished version. The owner has authorized the release, so record their operator-visible behavior first, as CONTRIBUTING requires: same-account Room discovery, delivery-lease renewal, and the versioned harness modes. Also record the release evidence and its limits. The tarball packed from the merged #61 commit is byte-identical to the reviewed artifact, and it was run as a production listener before publishing. That covers the live-listener and hosted-API gates #58 left open for serve. The serve-backed research modes still have no hosted or model-backed run, so the notes say so instead of implying it. Docs only: the packed tarball does not include docs/, so the artifact being published is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs(cli): correct the 0.1.16 notes on discovery backlog and leases An independent fact-check against 1921657 found two inaccurate sentences. A discovered Room does not "hand over nothing": its first mention carries everything since the kept place, and unlike a join it does not skip to the newest lines. And #56's lease keeps the Room showing the seat as replying past the old 20-minute window; it does not stop redelivery. Also state that #58's hosted no-model gate for the serve-backed modes was not run, add #61's known working-directory limit, and note that the driven binding must agree across an Instance's seats while --run uses its command. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Change and reason
Adds an executable, immutable harness-mode registry so research runs can name the exact coordination treatment they executed instead of relying on an ambiguous collection of flags.
goal run --harness-mode <name>@<version>now supports the six frozen research profiles:research/baseline@1research/mentions@1research/tasks@1research/both@1research/routing-only@1research/receipt-only@1The CLI rejects noncanonical names, conflicting low-level flags, compiled Rooms, agents without an explicit model, dirty source checkouts, and profile drift before the run starts.
episode.jsonrecords the mode, immutable semantic spec hash, fully resolved profile, clean commit/dirty state, package/entry/runtime-code hashes, image and pinned models, plus goal and opening-prompt hashes.The registry also reserves
runtime/serve-first@1andruntime/serve-all@1with their reviewed PR #48 provenance. They are explicitly unavailable in this build and fail before side effects; this PR does not claim that the unmerged serve-backed runner is supported. Mutable availability and re-verification metadata are excluded from the immutable mode hash, so enabling the exact reviewed implementation does not redefine@1.docs/harness-modes.mdis the catalog for delivery, wake, stop, flags, hashes, availability, and provenance. Existing ad hocgoal runflags remain compatible, and the root README is unchanged.Verification
pnpm check:export— 91 allowlisted filespnpm typecheckpnpm test— 429 passed, 1 skipped; 14 script tests passedpnpm test:package— installed-package smoke passed; tarball SHA-2560e8f413b8a89652f6b9c047ddf193da6781327b11e267958936e884fc67cee05No route, identity, or cursor behavior changes. Focused CLI and runner tests verify early validation, exact mode resolution, dirty-checkout handling, deterministic code provenance, unavailable runtime dependencies, stable semantic hashes, catalog consistency, and recorded episode metadata.
Compatibility and release
This adds one optional
goal runflag and preserves all existing low-level flags when it is omitted. Named research modes require board Rooms and an explicit model on every agent. Runtime mode names remain dependency-only until PR #48 is merged and wired in a later reviewed change.A source merge does not publish the package. Do not enable auto-merge or publish from PR CI.