Changelog

Added

• Intake now resolves an evidence-rich context, and the vulnerable multilspy dependency pin has been removed.

Added

• Added extraction of ontology code relationships.

Added

• Added support for provider-neutral intake writeback.

Added

• Added support for a collaborative intake scope.

Added

• Intake now supports grouped workflow bindings.

No user-visible changes.

Added

• You can now continue intake workflows directly from Slack.

Added

• API routes are now versioned.

Added

• Added deterministic intake workflow contracts.

Added

• Ontology contributor identities are now collected.

No user-visible changes.

Added

• You can now continue incomplete agent builds.

Fixed

• Projected serial PR claims are now accepted.

Fixed

• Required checks defined in rulesets are now correctly enforced.

Fixed

• Fixed an issue where the intake ontology could remain stale by refreshing it.

No user-visible changes.

Added

• Pull requests now support protected merge orchestration.

Added

• Intake runners now emit telemetry, including build provenance facts.

No user-visible changes.

Fixed

• Intake summaries are now written without requiring an app lookup.

Added

• Ontology refresh runs are now measured.

Fixed

• Correctly identify the GitHub App when writing back to issues.
• Added support for GitHub App issue writeback.

Fixed

• Ontology search repositories are now normalized.

Fixed

• Intake ontology is now scoped to installation repositories.

Added

• Added a private intake worker.

Fixed

• Build pull requests are now linked to their source issues.

Fixed

• Codex automation now uses supported flags.
• Resolved conflicting Codex sandbox flags.

Added

• Builds can now be orchestrated through huddles.

No user-visible changes.

Added

• Added semantic ontology routing.

Added

• Added an issue build agent runner.

Added

• Ingestion summaries are now written to issues.

Fixed

• Writeback now inspects all issue comments instead of only some.

Fixed

• Added support for the Anthropic ingestion schema.

Added

• GitHub issue events are now enriched during ingestion.

Added

• Release changelogs are now published to S3.

v0.6.0

Added

• Release changelogs can now be generated automatically with AI.

What's Changed

• Install LGTM v1.1.1 with Codex marketplace compatibility by @ike4est in https://github.com/sparepartslabs/spareparts-cli/pull/21

Full Changelog: https://github.com/sparepartslabs/spareparts-cli/compare/v0.5.2...v0.5.3

What's Changed

• Install LGTM for supported coding agents by @ike4est in https://github.com/sparepartslabs/spareparts-cli/pull/19
• Pin the published LGTM v1.1.0 digest by @ike4est in https://github.com/sparepartslabs/spareparts-cli/pull/20

Full Changelog: https://github.com/sparepartslabs/spareparts-cli/compare/v0.5.1...v0.5.2

Fixed

• Changelog entries are now published under the canonical changelog prefix.

Added

• Release changelogs are now published to S3.

v1.2.0

Added

• Plugin release changelogs can now be generated with AI.
• Plugin releases can now be created from semantic commits.

What's Changed

• chore: bump @vercel/ncc from 0.38.4 to 0.44.1 in the development group by @dependabot[bot] in https://github.com/sparepartslabs/spareparts-changelog/pull/7
• feat: add AI-generated public changelogs by @ike4est in https://github.com/sparepartslabs/spareparts-changelog/pull/8

New Contributors

• @dependabot[bot] made their first contribution in https://github.com/sparepartslabs/spareparts-changelog/pull/7

Full Changelog: https://github.com/sparepartslabs/spareparts-changelog/compare/v0.1.0...v0.2.0

What's Changed

• feat: add end-user changelog action by @ike4est in https://github.com/sparepartslabs/spareparts-changelog/pull/1
• ci: add protected Action release pipeline by @ike4est in https://github.com/sparepartslabs/spareparts-changelog/pull/2
• fix: reduce Dependabot update noise by @ike4est in https://github.com/sparepartslabs/spareparts-changelog/pull/6

New Contributors

• @ike4est made their first contribution in https://github.com/sparepartslabs/spareparts-changelog/pull/1

Full Changelog: https://github.com/sparepartslabs/spareparts-changelog/commits/v0.1.0

What's Changed

• chore(ec): install shared constitution workflow by @ike4est in https://github.com/sparepartslabs/spareparts-cli/pull/16
• fix(ec): sync canonical huddle status by @ike4est in https://github.com/sparepartslabs/spareparts-cli/pull/17

Full Changelog: https://github.com/sparepartslabs/spareparts-cli/compare/v0.5.0...v0.5.1

sp plugin

The new plugin command installs pinned, verified agent tools:

sp plugin list
sp plugin install lgtm --agent claude --dir .
sp plugin install lgtm --agent cursor --agent gemini
sp plugin install lgtm --all --dir .

Targets are claude, codex, cursor, copilot, gemini, and opencode. Existing files are preserved unless --force is passed.

The CLI verifies the LGTM archive before rendering each agent's native wrapper. Installation adds neither a Git hook nor a hidden memory edit. Without an agent target, sp plugin install lgtm retains the Codex marketplace path.

See Agent plugins for first-use behavior.

Added

• One agent-neutral command now installs for all six supported coding agents.
• On first use, LGTM can ask your agent to remember the quiz before its own pushes and pull-request creation, using a visible host mechanism only.
• The Codex marketplace path remains available alongside project-native targets.

sp plugin install lgtm --agent claude --dir .
sp plugin install lgtm --agent cursor --agent gemini --dir .
sp plugin install lgtm --all --dir .

Agent installation is separate from sp lgtm install, which adds a local Git hook.

Fixed

Workspace reconciliation preserves each huddle row's link to its repo-scoped specification directory.

Added

sp ec workspace sync reconciles shared huddles from the workspace API into the local .sp/huddles store while repo specifications remain repo-scoped.

Fixed

sp ec caps synchronized node status text so a detailed local status cannot make an otherwise valid workspace sync fail.

Fixed

The Homebrew release path keeps engineering-context commands dependency-light. Commands that do not call a model no longer inherit optional provider SDKs.

Added

sp ec can send a repository's current specification trajectory to its matching workspace node while keeping local Markdown authoritative.

sp ec

sp ec install adds engineering-context commands to a repository for the coding agents already in use there. It can detect one agent, target one by name, or install commands for every supported agent:

sp ec install
sp ec install --agent claude
sp ec install --all
sp ec install --dir path/to/repo

The installed workflow follows Spec Kit's specification, planning, task, and implementation structure while keeping /huddle as the cross-repository coordination command.

One .sp workspace

Generated context now lives under .sp/, replacing the earlier .blitz/ layout. Each repository keeps its own specs and command state. When installation targets a directory containing several repositories, the workspace root also gets a shared constitution and huddle store without receiving a pooled repository scaffold.

The installer carries over the existing ignore behavior: generated .sp state and templates are ignored, while .sp/memory/constitution.md remains trackable. Existing blanket .sp/ ignore rules are preserved and reported rather than silently rewritten.

Huddles and GitHub Projects

A workspace can use a GitHub Project as its huddle store. Configuration and manual synchronization are available from the command line:

sp ec project configure https://github.com/orgs/example/projects/1 --dir ..
sp ec project status --dir ..
sp ec project sync ../.sp/huddles/001-example/huddle.md --dry-run
sp ec project sync ../.sp/huddles/001-example/huddle.md

/huddle prefers installed GitHub MCP tools that can manage Projects and falls back to gh project through the CLI. Every synchronized draft item carries a stable huddle-path marker, so another sync updates the same item instead of creating a duplicate. The Markdown document remains authoritative if remote synchronization is unavailable.

Guided setup

Two commands make integration requirements visible before a huddle needs them:

sp ec doctor
sp ec project setup

They report the configured huddle store, GitHub CLI authentication, and visible Linear MCP or API configuration. Missing capabilities include the next command or configuration step needed to enable them. GitHub Projects is fully synchronized in this release; Linear setup and detection establish the path for its work-management integration.

Install

brew upgrade sparepartslabs/tap/spareparts-cli
pip install --upgrade spareparts-cli

See the Spare Parts CLI documentation for installation options and the complete sp ec workflow.

sp is the Spare Parts command line. Modules are imported lazily, so the module named on the command line is the only one loaded: running one never pays for another's imports, and one with a broken dependency cannot take the rest of the command line down with it.

sp lgtm

Generates a few multiple-choice questions about a range of commits and asks them in your terminal, before the code leaves your machine.

sp lgtm                       # what this branch adds since it left main
sp lgtm main...feature/x      # someone else's branch, before you merge it
sp lgtm --staged              # what is staged
sp lgtm --dry-run             # what it would ask about, no model call

Press ? on any question to print the hunk it came from. Wrong answers are never a failure: you are told which files to look at again, and it re-asks as many times as you like.

It reads .github/lgtm.yml, the same file the GitHub Action reads, so a repository is configured once. Fields that only mean something to the Action are accepted and ignored.

As a git hook

sp lgtm install                       # pre-push, advisory
sp lgtm install --hook pre-commit     # earlier, and once per commit
sp lgtm install --blocking            # wrong answers stop the push

pre-push is the default, and advisory is the default. A hook that costs a minute and can stop you is one you delete within a week. Even with --blocking, only a wrong answer blocks: could-not-ask, meaning no API key, a vendor outage, or nothing quizzable, never costs you a push.

The hook quizzes exactly what you are about to push, reading the ref range git puts on stdin, and falls back to what the branch adds since it left the default branch when the remote has never seen it. It skips itself silently where there is no terminal, such as a rebase, a GUI client, or CI. sp lgtm install honours core.hooksPath and refuses to overwrite a hook it did not write.

Vendors

No vendor is the assumed one. On pip each SDK is an extra, so you install the client for the vendor you actually call and nothing else.

Name no vendor and sp uses whichever key is set. The defaults were chosen by listing each vendor's models against a live key rather than from memory.

Split the proposer and the verifier across two vendors if you have the keys for it. A model asked to find fault with its own reasoning mostly does not, so this is the strongest arrangement available, and it costs one extra vendor's tokens and almost no extra wall clock:

sp lgtm --provider anthropic --verifier openai

Install

brew install sparepartslabs/tap/spareparts-cli
pip install "spareparts-cli[anthropic]"     # or [openai], [gemini], [all]

The Homebrew formula carries all three vendors and brings its own Python rather than touching the one you develop against. Dependencies are otherwise thin on purpose: everything the terminal rendering needs is in the standard library.

Exit codes

Treat only 1 as a failure. A tool that cannot run must not block a commit.

Fixed

• The action published as LGTM, a name generic enough to be hard to find and easy to mistake for someone else's. It is Spare Parts LGTM now.

Listing metadata only, with no change to behaviour, inputs or outputs. Workflows pinned to sparepartslabs/spareparts-lgtm@v1 already point at it.

LGTM asks the reviewer two or three questions about the pull request they just approved, generated from the diff and reported as a check run. It has no opinion on whether the change is good, only on whether anyone read it.

The shape it ships in

Two workflow files and two secrets. There is no server to run, no database to provision, and no checkout: the diff and the configuration are both read through the API. Quiz state lives in the comment and the check run, configuration in .github/lgtm.yml.

The install is split in half on purpose. lgtm-collect.yml listens for pull_request_review with permissions: {} and does nothing but write the event to an artifact, because that event carries no secrets on a pull request from a fork. lgtm.yml is the privileged half and picks the relayed event up through workflow_run, treating the artifact as untrusted.

Questions

Three stages run on the diff, after a file screen that makes the common skip cost nothing.

• Propose. One call writes candidates about what the change does: what a new guard prevents, which edit touches existing rows, what an error path now returns.
• Verify. An independent call that never saw the proposer's reasoning is told to refute each candidate. It survives only if that call agrees the answer is right and that someone who skipped the diff could not guess it.
• Ground. The cited file and hunk are checked against the parsed diff in code, and the options are checked for structural tells.

Everything is conservative in one direction. A quiz with no survivors is not posted and the check concludes neutral. The answer key is never published: the comment carries a keyed hash per correct option, so an edited quiz is reissued rather than graded.

Asking a question back

Mention @lgtm at the start of a line to get background questions answered, with web search where a source helps. Questions that ask LGTM to read the pull request for you are declined, and you get a pointer at which file answers it instead. Write access only, rate limited per asker per pull request, and answerQuestions: false turns it off.

Vendors

Anthropic, OpenAI and Gemini, behind one provider seam, and any of the three can run the web search. Each repository supplies its own model key, so adoption costs the adopter.

Enforcing

Off by default. enforce: true in .github/lgtm.yml holds a merge until the quiz is answered. LGTM can never mark a review as failed, and unlimited attempts are the point rather than a concession.