v0.6.1 - Security audit refresh

Highlights

• Clarified the public runtime boundary: bundled retrieval and runner-adaptation
  scripts accept data-only formats, use release-fixed imports, and neither
  launch the optional upstream runner nor contact the network.
• Replaced three type-narrowing assertions in the runner adapter with explicit
  fail-closed errors, preserving required-input invariants under optimized
  Python without changing command-line interfaces, schemas, lifecycle
  transitions, or artifact behavior.
• Refreshed the published bundle content so registries and external security
  providers can index a distinct v0.6.1 snapshot.

Release verification

• 87 unit tests passed; the 6-test runner-adapter suite also passed under
  optimized Python.
• 120/120 isolated offline contract-canary runs passed across 12 cases with
  zero illegal transitions.
• Skill validation, pinned skills CLI discovery/rendering, public repository
  discovery, Bandit with zero findings, dangerous dynamic-loader pattern
  scanning, and git diff --check passed.

Security context

• The removed v0.5.0 embedding-provider and sibling-module loaders remain
  absent from every published script.
• The public safety contract now states the executable boundary directly rather
  than relying on readers to infer it from the individual backend and adapter
  descriptions.
• skills.sh audit results are generated externally and may remain unchanged
  until the new bundle is indexed and partner audits are rerun. The stale-audit
  report is tracked in
  vercel-labs/skills#1673.

Install or update

npx skills add lizhouai/storm-research-skill

For an existing project-local installation:

npx skills update storm

v0.6.0 - Security hardening

Highlights

• Removed the experimental in-process embedding-provider loader, eliminating
  the path that could import and execute a user-selected Python module or file.
• Limited bundled retrieval to Agent-ranked host input and deterministic,
  zero-dependency lexical search over a user-provided local corpus.
• Replaced the runner adapter's dynamic sibling-module loader with a fixed
  import and changed upstream dependency probing to distribution metadata only.
• Aligned the skill contract, public documentation, contributor guidance, and
  automated regression checks with the new data-only retrieval boundary.

Upgrade note

The experimental embedding backend and its --embedding-provider, --model,
--provider-version, and --fallback options have been removed. Existing
workflows should use host for rankings produced by the current Agent or
runner, or lexical for deterministic retrieval inside a local corpus.

Release verification

• 87 unit tests passed.
• 120/120 isolated offline contract-canary runs passed across 12 cases with
  zero illegal transitions.
• Skill validation, Python compilation, pinned skills CLI discovery/rendering,
  dangerous dynamic-loader pattern scanning, and git diff --check passed.

Known limitations

• Retrieved sources and external runner output remain untrusted data and still
  require the existing prompt-injection, citation, and artifact-validation
  boundaries.
• skills.sh security results are generated externally and may continue to show
  the previous release until the repository is re-indexed.
• Co-STORM remains a prompt-native preview with simulated participants and does
  not claim upstream CoStormRunner parity.

Install or update

npx skills add lizhouai/storm-research-skill

For an existing project-local installation:

npx skills update storm

v0.5.0

Highlights

• Added guarded experimental retrieval routing for Agent-ranked evidence, local lexical corpora, and explicitly configured embedding providers, with deterministic traces and corpus-boundary enforcement.
• Added an experimental zero-dependency adapter for importing compatible official Classic STORM outputs into the existing guarded artifact lifecycle without installing or executing the upstream runner.
• Improved intent routing so users describe the research outcome or inputs they have; internal backend and adapter labels are inferred rather than exposed as user-facing concepts.
• Expanded the public documentation with concise usage examples, capability boundaries, recovery behavior, and accurate experimental labels while preserving the storm skill name and install path.

Release verification

• 87 unit tests passed.
• 120/120 isolated offline contract-canary runs passed across 12 cases with zero illegal transitions.
• Skill metadata validation, Python compilation, pinned skills CLI discovery/rendering, and installed-bundle smoke checks passed.
• Documentation links, repository URLs, and the project-local install path were checked against the release tree.

Experimental capability boundaries

• Retrieval routing and official Classic output import are experimental in v0.5.0.
• Embedding mode requires an explicit provider command, model, and provider version; the skill does not discover or install providers.
• The official adapter supports stable knowledge-storm output contracts in >=1.1.1,<1.2; it imports completed output only and does not bundle, install, or execute the upstream runner.
• Adapter tests use a compatible knowledge-storm 1.1.1 fixture because the upstream package is not installed in the release-validation environment.
• Co-STORM remains a prompt-native preview with simulated participants and does not claim upstream CoStormRunner parity.
• Offline fixture canaries validate workflow contracts, not real-model research quality.

Install or update

npx skills add lizhouai/storm-research-skill

For an existing project-local installation:

npx skills update storm

v0.4.1 - Guarded verification hardening

Highlights

• Hardened Classic STORM completion with staging-first publication, strict citation and artifact evidence, SHA-256 rechecks, publication receipts, and rollback for recoverable publication failures.
• Added persistent Co-STORM turn validation with a versioned schema, stable participant and source mappings, idempotent hash-linked logs, and lifecycle gates for warm start and final reporting.
• Synchronized README, CONTRIBUTING, examples, CI, forward-eval fixtures, and installed bundle checks with the guarded runtime contract.
• Preserved the public skill interface: skills/storm, name storm, and $storm remain unchanged.

Release verification

• 66 unit tests passed.
• 120/120 isolated offline contract-canary runs passed across 12 cases with zero illegal transitions.
• Pinned and latest skills CLI discovery/render checks passed, including the public owner/repository install path.
• GitHub Validate and Pages deployment passed for commit 0504f88.

Capability boundaries

• Co-STORM remains a prompt-native preview and does not claim upstream CoStormRunner parity.
• Offline fixture canaries validate the contract, not real-model research quality.
• Classic validation does not mechanically verify the semantic quality of Co-STORM reports.
• B5 retrieval upgrades and B6 local-runner adapters remain intentionally out of scope.

v0.4.0 - Guarded STORM runtime

Highlights

• Added a zero-dependency guarded runtime with versioned run state, atomic checkpoints, legal phase transitions, recovery, event hashes, and explicit completion state.
• Added structural artifact validation and citation auditing so Classic STORM cannot complete with malformed files, dangling sources, or unsupported claim mappings.
• Split the workflow into focused Classic, Co-STORM, Local Runner, artifact, safety, and state-schema references while preserving the compatibility index.
• Made guarded execution the default for file-producing Classic and Local Runner requests when Python is available.
• Expanded the executable offline forward-eval contract to 12 cases, including Co-STORM warm start, visible follow-up, chat conclusion, and report-only persistence.
• Documented the unbounded Co-STORM interaction loop, final-report trigger, request-scoped file output, and an end-to-end RAG technology report example.
• Added an upload-ready 1280x640 Social Preview that makes the Fact Researcher, Specialist, Respondent/Critic, and Moderator roles visible.
• Renamed the public repository to lizhouai/storm-research-skill; install with npx skills add lizhouai/storm-research-skill.

Release verification

• 55 unit tests passed.
• 120/120 deterministic offline contract-canary runs passed with zero illegal transitions, 120 fresh subprocesses, and 120 isolated work directories.
• skills@1.5.15 discovered and rendered the full bundle locally and from the public owner/repository path.
• GitHub Validate and Pages deployment passed on main.
• Both public Classic and Co-STORM HTML examples returned 200 text/html from the renamed Pages site.
• Social Preview verified at 1280x640 and 377,429 bytes.

Capability boundaries and known gaps

• Co-STORM remains a prompt-native preview with simulated, visibly attributed participants. This release does not bundle or execute the upstream CoStormRunner or independently running expert agents.
• Persistent Co-STORM state guards outer lifecycle transitions and checkpoint integrity. The Classic artifact validator does not mechanically verify Co-STORM mind-map or report contents; source and citation support still require review.
• Offline fixture canaries validate the evaluation contract, not real-model research quality. The optional real-Codex-host adapter remains non-blocking, and the previously observed local CLI/account model-compatibility gap is not a release gate.
• B5 retrieval upgrades and B6 local-runner adapters are intentionally not included in this release.

v0.3.0

This release adds a prompt-native Co-STORM preview while keeping classic STORM artifact generation as the default.

Highlights:

• Adds visible, role-attributed roundtable turns, choice-first steering, cited mind-map updates, moderator cadence, and checkpoint/recovery rules.
• Documents safety boundaries for untrusted sources, local runners, output conflicts, remote side effects, and checkpoint reloads.
• Adds eval fixtures, dependency-free validation, regression tests, and GitHub Actions coverage for pinned and latest skills CLI discovery and rendering.
• Clarifies install/update scope, cross-agent compatibility, and that this repository does not bundle the upstream executable Co-STORM engine or DSPy modules.

Full Changelog: v0.2.1...v0.3.0

v0.2.1

Docs: add a contributing guide, clarify that Co-STORM-style workflows are still in development, and remove duplicated artifact wording from the README. The v0.2.1 tag points to the latest main commit for this release.

v0.2.0

Changes

• Make non-interactive STORM research emit the standard four artifact files by default.
• Default unspecified output format to HTML.
• Default unspecified output path to .results//.
• Add UTF-8 and Windows file hygiene guidance for generated artifacts.
• Document the v0.2.0 artifact defaults in README and OpenAI metadata.

Verification

• Python structural checks passed for repository and installed skill copies.
• git diff --check passed before release.

v0.1.0

Initial public release of the STORM Research Agent Skill.\n\nHighlights:\n- Packages the STORM research workflow as a SKILL.md-based Agent Skill.\n- Supports classic STORM research reports and Co-STORM-style interactive exploration.\n- Adds upstream attribution to stanford-oval/storm.\n- Provides a simple install command: npx skills add lizhouai/storm.\n- Releases this repository under the MIT License.