feat: model invokable workflow skills - #93
Open
tkstang wants to merge 90 commits into
Open
Conversation
Removing `disable-model-invocation: true` lets orchestrator skills (notably `oat-project-document`) chain this skill under Claude's harness, matching the behavior Codex already exhibits by ignoring the flag. Tightens the description so Claude does not fire it speculatively outside the doc-sync flow. Bumps skill version 1.0.0 -> 1.1.0 and lockstep-bumps the five public packages 0.0.50 -> 0.0.51 per AGENTS.md release policy.
* feat(skills): make end-of-lifecycle skills model-invokable with explicit-ask gating Removes `disable-model-invocation` from `oat-project-document`, `oat-project-pr-final`, `oat-project-summary`, and `oat-pjm-add-backlog-item` so the model can run them when the user confirms a previously offered step. Each description now leads with the explicit-ask trigger phrasings and a "do NOT auto-invoke" clause naming the false-positive trigger to suppress, so discovery still filters out unsolicited invocations. Bumps the lockstep public package versions per the AGENTS.md guardrail. * refactor(skills): tighten gating descriptions and broaden trigger to user confirmation Replaces "explicitly asks" with "requests or confirms" so the description matches the dominant real-world path: bot offers to run the skill at end of implementation, user replies "yes". Each example list now includes "or confirms a previously offered <step>" so the model recognizes a bare confirmation as a valid trigger. Drops the redundant "wait for an explicit user signal" tail clause; the lead phrasing and example list already cover that ground. * chore(cli): untrack bundled CLI assets These files were tracked by accident before `packages/cli/assets/` was added to .gitignore — they are derived artifacts regenerated by `packages/cli/scripts/bundle-assets.sh` during `pnpm build`, and the release pipeline (`tools/release/validate-public-packages.ts`) builds each public package before packing, so the published tarball always ships fresh content. Tracking them created two problems: silent drift between checked-in versions and what `pnpm build` produced, and `git add` rejecting routine updates because the path is ignored. Untracking removes both.
…ing to oat-project-spec (#77) * fix(skills): rewrite PR description after archive in oat-project-complete After oat-project-complete archives a project locally (per pre-merge archive flow), the open PR's description still pointed at the active artifact paths (plan.md, implementation.md, etc.) — those files were moved to a gitignored archive location and the Step 10 commit removed them from the branch, so every blob link in the PR body 404'd while the PR was still open. oat-project-complete now captures WAS_PR_OPEN_AT_START before any state mutation, regenerates the PR artifact with archive-aware References when SHOULD_ARCHIVE is true (drops bullets for artifacts about to be local-only, adds a project-record link to the summaryExportPath copy on the current head branch), and a new Step 11.5 syncs that body to the open PR via `gh pr edit`. The project-record link targets {BRANCH} rather than {BASE_BRANCH} because the summary export is committed on the feature branch in Step 10 and only reaches the base branch after merge. Contract test in review-skill-contracts.test.ts pins the new behavior and guards against regressing the project-record link to {BASE_BRANCH}. * fix: stop recommending oat-project-spec as a default next step Requirements confirmation was folded into oat-project-design, but the state-dashboard router, the new-project scaffold, and several router skills still treated `spec` as a discrete phase and pushed users at oat-project-spec right after discovery. Newly scaffolded projects opened with `Phase: spec (in_progress)` and `Recommended next: oat-project-spec`, undoing the consolidation. Source-of-truth changes: - Scaffold spec-driven projects with HiLL checkpoints ['discovery', 'design'], reword the Spec artifact bullet to indicate oat-project-design authors it inline, and update the next-milestone text to reference the design phase. - Route spec-driven discovery:complete to oat-project-design (not oat-project-spec). Legacy projects whose state.md still reads spec:in_progress also route to design as graceful migration. - Map the HiLL phase->skill table so a pending `spec` checkpoint on a legacy project routes to oat-project-design. Skill updates: - oat-project-progress (1.2.3 -> 1.2.4): HiLL recommendations, spec-driven routing matrix, and the help-text inventory now point at oat-project-design and frame oat-project-spec as an opt-in standalone. - oat-project-next (1.0.3 -> 1.0.4): HiLL phase->skill map and the spec-driven boundary-tier routing table all route spec entries to oat-project-design, with a clarifying note about legacy projects. oat-project-spec itself is preserved as a deliberate opt-in path. oat-project-design already reuses any existing spec.md, so the standalone skill is still wired into the consolidated flow. Tests: - Added two generate.test.ts cases pinning the new routing (spec-driven discovery:complete -> design; legacy spec:in_progress -> design). - Updated the scaffold.test.ts spec-artifact assertion to match the new wording. Lockstep public package bump 0.0.67 -> 0.0.68 (third content stage on this branch). * chore: gitignore .oat/repo/reviews and drop the in-PR review artifact Treat repo-level review artifacts under .oat/repo/reviews as local-only scratch, the same way archived review artifacts and brainstorm/idea artifacts are already handled. Add the directory to .gitignore and to the localPaths list in .oat/config.json so the oat-project-pr-final References policy stops linking to paths that will not exist on the remote. Also drop the ad-hoc review artifact that landed in commit 1108e06 (it documents the review of that commit and was never meant to ship as repo content). Pre-existing review files committed before this rule remain tracked as historical content. The new gitignore rule only affects future adds — already-tracked files keep tracking as-is.
…ntions, not file count (#86)
tkstang
force-pushed
the
feat/model-invokable-workflow-skills
branch
from
June 4, 2026 23:50
81e4929 to
6e3e2f0
Compare
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
1 participant
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.
Summary: skill-automation-and-review
Overview
This project reduced OAT lifecycle friction by making generated artifacts harder to consume unchecked, making review artifacts easier to resolve from natural-language workflows, and expanding model-invokable lifecycle skills with explicit offer-and-confirm gates.
What Was Implemented
workflow.autoArtifactReview.planandworkflow.autoArtifactReview.analysis, both defaulting totrue, so generated plans and generated docs/agent-instructions analysis artifacts run bounded review loops before implementation or apply workflows consume them.oat-reviewerwithartifact: planandanalysis: docs|agent-instructionsstructured review modes. The reviewer stays stateless; authoring skills own the retry loop and artifact rewrites.oat-project-plan-writingowns the canonical auto artifact-review loop contract.oat-project-plan,oat-project-quick-start, andoat-project-import-planinvoke it for plans;oat-docs-analyzeandoat-agent-instructions-analyzeinvoke it for analysis artifacts.oat review latestto scan project and ad-hoc review locations and return the newest artifact by frontmatter timestamp. A final-review fix added lifecycle recency for same-day ties sofinaloutranks phase reviews and higher phase/task scopes outrank lower ones.oat-project-review-provide,oat-project-review-receive,oat-project-discover, andoat-project-progressmodel-invokable for explicit asks only. Each skill documents its gate and prevents silent workflow jumps.oat-project-quick-startnow completes discovery in the lightweight-design path before plan generation, preventing quick-mode projects from reaching a complete plan withdiscovery.mdstillin_progress.0.1.18.Key Decisions
oat-revieweragent rather than creating a separate artifact-review agent.oat_orchestration_retry_limitinstead of introducing another limit key.Design Deltas
p01-t01expanded beyond its initial file list to update config resolver and config command support required byoat config get workflow.autoArtifactReview.*.p03updated the quick-start skill version contract tests alongside planned quick-start skill version bumps.p05needed a review-fix commit to alignoat-project-review-provideStep 0 with the advertised active-project or explicit-target gate.p06and final review both found config-precedence wording drift. Docs and config catalog now state thatworkflow.autoArtifactReview.*has no environment aliases and resolves throughlocal > shared > user > default.oat review latest.Verification
p01throughp06passed.pnpm build,pnpm lint,pnpm format,pnpm type-check,pnpm test,pnpm release:validate, andgit diff --check.oat review latest --project ... --json,config describe workflow.autoArtifactReview.plan,config describe workflow.autoArtifactReview.analysis,git diff --check, andpnpm release:validate.