Releases: sockulags/agent-os
Release list
agent-os v0.11.0
The proportional-testing release: protect changed behavior without turning every small fix into a large test-generation exercise.
- New automatic
proportional-testingdiscipline requires a realistic, previously unprotected regression before adding another test. - Prefers existing and table-driven tests, observable contracts, and the cheapest sufficient test level.
- Integrates focused bug-fix coverage into
deliver-workwhile preserving required repository, CI, and release checks.
This is a GitHub-only release. npm publication is intentionally excluded; @sockulags/agent-os remains at 0.9.1 on npm.
agent-os v0.10.0
Simplification tools
simplifierreduces unnecessary code, dependencies, abstractions, and solution layers while preserving required behavior and readability.simplifier-auditscans a repository for high-confidence simplification opportunities without applying changes.simplifier-reviewreviews the current diff for unnecessary complexity and now runs as part ofdeliver-workbefore a candidate is frozen.list-skillsreports the installed Agent OS skills and their invocation boundaries.- Added the documented Codex custom-instruction experiment, including mutation, rollback, observed results, and limitations.
This publication is the GitHub tag and GitHub Release. The npm package remains at 0.9.1 by request.
agent-os v0.9.1
Portable direct-installer correction.
- installs all packaged Agent OS skills directly for Claude Code, Codex, or both without requiring either host CLI
- records managed directories and preserves unrelated skills during updates
- syncs global policy through Node with malformed-marker preflight
- retains native marketplace installation behind
--method plugin - skips Git refresh for local Codex marketplaces in optional plugin mode
Published on npm as @sockulags/agent-os@0.9.1.
See the getting started guide and changelog.
agent-os v0.9.0
Initial guided npm installer release.
- adds
npx @sockulags/agent-os installandupdate - adds guided platform, scope, and policy choices
- publishes the scoped npm package and validates package metadata with the plugin manifests
This version delegated installation to host marketplace CLIs and is superseded by v0.9.1, which installs packaged skills directly and fixes local Codex marketplace updates.
See the changelog for details.
agent-os v0.8.0
Guide-me workflow release.
- adds
guide-meas the on-ramp from a vague desire to approved shaping or charting - adds
understand-work,explain-work,notice-lesson, andrecord-lesson - expands workflow documentation, troubleshooting, examples, and trigger eval coverage
- keeps workflow invocation explicit while allowing approved workflow handoffs
See the changelog for details.
agent-os v0.7.0
Implementation-ready issues
shape-worknow completes by creating or reusing concrete implementation issues on the configured planning surface.- Each issue carries stable identity, readiness, outcome, boundaries, acceptance, ground truth, dependencies, evidence, delivery target, and open risks.
- Research, grilling, and prototypes remain decision evidence; implementation issues are the durable delivery-ready output.
Developer-owned execution
- Multiple issues no longer imply
batch-work. - Any ready issue can be selected for
deliver-work. batch-workconsumes an existing dependency-mapped issue graph only after an explicit developer request for integrated batch execution.dispatch-nextselects one ready issue unless a batch was explicitly requested.
Lifecycle safeguards
- A chart-work handoff moves a branch to shaping, not delivery-ready.
- Selected shaping branches keep the map open until issues and a frontier exist.
deliver-workstops before mutation when a target contains several independently closable outcomes or delivery boundaries.- Project policy setup now captures implementation-issue readiness conventions.
Evidence
- Repository validator passed.
- All validator red cases passed, including new issue-materialization, routing, and decomposition failures.
- Batch manifest tests passed.
- VitePress production build and GitHub Pages deployment passed.
- A sealed read-only Codex 0.146.0 forward test produced two ready issues, one dependency-blocked verification issue, reconciled the handoff and map, and exposed the correct frontier without choosing batch. Actual tracker mutation and retry idempotency remain unmeasured.
agent-os v0.6.2
Independent review gate
deliver-worknow requires independent review for material changes.- Review may be skipped only when every small-fix condition holds: one localized supported defect, no new capability or public/sensitive boundary, and direct regression evidence.
- Reviewer identity must come from a real launch result in the current run. Empty waits, invented reviewer labels, self-review, and unreturned reviewers do not count.
- Supported findings require a fix, targeted re-review, and fresh final verification.
- Integrated
batch-workcandidates use the same material review gate.
Evidence
- Repository validator and all red cases passed.
- Batch manifest tests passed.
- VitePress production build passed.
- Fresh installed-plugin trigger matrix: 26/26 measured cases passed.
- Forward tests passed for a localized fix, a material feature with a real reviewer, and an MCP auth/public-schema/external-write change. The MCP reviewer found a missing/empty-token bypass; the implementer fixed it and a targeted re-review returned no findings.
Known limitation
A raw CLI host that exposed wait but no reviewer launch tool did not obey the required pre-mutation stop and invented a reviewer identity. The contract now rejects that behavior, and the published Evals page records the case as FAIL. Material delivery is not claimed as independently reviewed in that host.
agent-os v0.6.1
Evidence-backed documentation
- Adds real-session transcript demos for all automatic disciplines.
- Publishes measured trigger accuracy, positive and negative outcomes, and forward-test history.
- Traces one feature through chart-work, shape-work, batch-work, and deliver-work.
- Fixes rendered last-updated timestamps and page-specific Open Graph metadata.
- Rejects drift between skill frontmatter, documentation pages, and overview tables.
This release changes documentation, validation, and version metadata. Skill sources are identical to v0.6.0.
Verification
- Repository contracts: pass
- Validator red-case suite: pass
- Batch manifest hashing: pass
- VitePress production build: pass
- GitHub Validate workflow: pass
- GitHub Pages build and deploy: pass
- Published Evals, Guide, and verify-before-done pages: HTTP 200 with expected release content
agent-os v0.6.0
What changed
- The developer request is now the authority boundary: implementation requests authorize in-scope repository edits without a redundant approval checkpoint.
deliver-workis contract-driven around outcome, boundaries, ground truth, and the requested delivery target.- Routine work no longer carries a procedural review ledger; independent review remains proportional for explicitly requested or unusually risky work.
- Batch, chart, shape, dispatch, initialization, diagnostics, scope, verification, and skill-writing flows have been simplified around agent judgment inside explicit bounds.
- Batch hashes now protect task and manifest drift rather than model approval state.
- Documentation, validators, eval cases, and both Claude Code and Codex plugin manifests are synchronized at 0.6.0.
Validation
- Repository contract validator and its red-case suite
- Batch manifest hash tests
- VitePress production build
- Pull request CI on the exact release commit
agent-os v0.5.0
Highlights
- Adds manual batch-work orchestration for several approved, decision-complete units.
- Freezes batch plan and task definitions behind one explicit manifest-hash approval checkpoint.
- Uses isolated deliver-work workers with local commit + receipt boundaries.
- Adds canonical runtime state, exact stale-receipt rejection, bounded retries, serial integration and fresh aggregate verification.
- Keeps PR, merge and cleanup authority as separate frozen + current two-key checks.
- Routes chart-work decisions through shape-work, then one ready unit to deliver-work or several to batch-work.
Verification
- Repository validator and 19 isolated red cases
- Batch manifest parser/hash/CLI adversarial suite
- Fresh-agent plan, resume, routing, fallback, authority and non-trigger cases
- VitePress documentation build
- GitHub Validate and Docs workflows on the release merge commit