Skip to content
Back to search

chrisbanes-skills

MarketplaceWorks with Claude CodeGitHub

Skills for Kotlin, Jetpack Compose, and Android development

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/chrisbanes/skills with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/chrisbanes/skills'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/chrisbanes-skills/badge.svg)](/listings/chrisbanes-skills)

HTML

<a href="/listings/chrisbanes-skills"><img src="/api/listings/chrisbanes-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

With the [skills CLI](https://skills.sh):

npx skills add chrisbanes/skills

Or install as a Claude Code plugin:

/plugin marketplace add chrisbanes/skills
/plugin install chrisbanes-skills@chrisbanes-skills

Or install as a Codex plugin:

codex plugin marketplace add chrisbanes/skills --ref main
codex plugin add chrisbanes-skills@chrisbanes-skills

Or install as an OpenCode V2 plugin (2.0.22 or newer):

{
  "plugins": ["chrisbanes-skills@git+https://github.com/chrisbanes/skills.git"]
}

OpenCode V1 1.18.29 or newer uses `plugin` instead of `plugins`. See [`.opencode/INSTALL.md`](.opencode/INSTALL.md) for details.

External skill providers

Most skills in this repository are self-contained. The workflows below compose with skills from other repositories; installing `chrisbanes/skills` does not install them, and the workflows never install them implicitly.

| Consumer | Requirement | Provider | Source and install | |---|---|---|---| | `deliver-spec` GitHub readiness | Conditional | `triage` | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill triage` | | [`implement-with-subagents`](skills/implement-with-subagents/SKILL.md) behavior tasks | Conditional | `tdd` at approved test seams | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill tdd` | | [`run-github-project`](skills/run-github-project/SKILL.md) explicit/repository-required test-first work | Conditional | `tdd` | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill tdd` | | `run-github-project` Todo triage | Conditional | `triage` | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill triage` | | `run-github-project` Wayfinder lane | Conditional | `wayfinder`, `research` | [Matt Pocock's skills](https://github.com/mattpocock/skills): use `--skill wayfinder` or `--skill research`; see the [workflow provider matrix](skills/run-github-project/references/workflow-providers.md) |

Delivery and parallel implementation use one fresh independent read-only reviewer against the approved source and repository standards; no external review skill or issue-tracker setup is required. Reviews include the production paths and boundary cases examined and material uncertainty. The authoritative [behavioral review and repair procedure](references/behavioral-review.md) covers related boundary repairs, repeated external findings and browser diagnostics. `npm run references:sync` supplies identical local copies to its consumers; edit the authoritative source, not the copies. The same mechanism supplies [to-plan's boundary-validation rule](skills/to-plan/references/boundary-validation.md) to implementation consumers.

Review mode in `implement-with-subagents`, and review or setup mode in `run-github-project`, do not require these external skills. See the [`run-github-project` provider matrix](skills/run-github-project/references/workflow-providers.md) for its lane-specific fallback and blocking behavior.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 10 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. {
  2. "name": "chrisbanes-skills",
  3. "owner": {
  4. "name": "Chris Banes",
  5. "url": "https://github.com/chrisbanes"
  6. },
  7. "plugins": [
  8. { "name": "chrisbanes-skills", "source": "./" }
  9. ]
  10. }
README.md · README · 433 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. ![Abstract illustration of a modular Kotlin and Jetpack Compose toolkit](docs/assets/skills-header.webp)
  2. # Skills
  3. A set of skills for Kotlin, Jetpack Compose, Android development, and grounded
  4. writing.
  5. The repository is also a portable [Agent Plugins](https://agent-plugins.org/)
  6. v1.0.0 package. Conforming clients discover the root [`plugin.json`](plugin.json)
  7. and the immediate skill directories under [`skills/`](skills/).
  8. ## Install
  9. With the [skills CLI](https://skills.sh):
  10. ```
  11. npx skills add chrisbanes/skills
  12. ```
  13. Or install as a Claude Code plugin:
  14. ```
  15. /plugin marketplace add chrisbanes/skills
  16. /plugin install chrisbanes-skills@chrisbanes-skills
  17. ```
  18. Or install as a Codex plugin:
  19. ```
  20. codex plugin marketplace add chrisbanes/skills --ref main
  21. codex plugin add chrisbanes-skills@chrisbanes-skills
  22. ```
  23. Or install as an OpenCode V2 plugin (2.0.22 or newer):
  24. ```json
  25. {
  26. "plugins": ["chrisbanes-skills@git+https://github.com/chrisbanes/skills.git"]
  27. }
  28. ```
  29. OpenCode V1 1.18.29 or newer uses `plugin` instead of `plugins`.
  30. See [`.opencode/INSTALL.md`](.opencode/INSTALL.md) for details.
  31. ### External skill providers
  32. Most skills in this repository are self-contained. The workflows below compose
  33. with skills from other repositories; installing `chrisbanes/skills` does not
  34. install them, and the workflows never install them implicitly.
  35. | Consumer | Requirement | Provider | Source and install |
  36. |---|---|---|---|
  37. | [`implement-with-subagents`](skills/implement-with-subagents/SKILL.md) behavior tasks | Conditional | `tdd` at approved test seams | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill tdd` |
  38. | [`run-project`](skills/run-project/SKILL.md) explicit/repository-required test-first work | Conditional | `tdd` | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill tdd` |
  39. | `run-project` Todo triage | Conditional | `triage` | [Matt Pocock's skills](https://github.com/mattpocock/skills): `npx skills add mattpocock/skills --skill triage` |
  40. | `run-project` Wayfinder lane | Conditional | `wayfinder`, `research` | [Matt Pocock's skills](https://github.com/mattpocock/skills): use `--skill wayfinder` or `--skill research`; see the [workflow provider matrix](skills/run-project/references/workflow-providers.md) |
  41. Delivery and parallel implementation use one fresh independent read-only reviewer
  42. against the approved source and repository standards; no external review skill
  43. or issue-tracker setup is required. Reviews include the production paths and
  44. boundary cases examined and material uncertainty. The authoritative
  45. [behavioral review and repair procedure](references/behavioral-review.md) covers
  46. related boundary repairs, sensitive fixture calibration before expansion,
  47. rendering-consumer tracing, repeated external findings and browser diagnostics.
  48. `npm run references:sync` supplies identical local copies to its consumers;
  49. edit the authoritative source, not the copies. The same mechanism supplies
  50. [to-plan's boundary-validation rule](skills/to-plan/references/boundary-validation.md)
  51. to implementation consumers.
  52. Delivery and planning select affected existing tests through shared-structure
  53. consumers and structural audits. Unsettled fixtures establish one sensitive case
  54. before expanding the acceptance matrix; established fixtures reuse valid evidence.
  55. Review mode in `implement-with-subagents`, and review or setup mode in
  56. `run-project`, do not require these external skills. See the
  57. [`run-project` provider matrix](skills/run-project/references/workflow-providers.md)
  58. for its lane-specific fallback and blocking behavior.
  59. ## Skills
  60. ### Start here
  61. - Working on Compose state or effects? Start with [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md).
  62. - Investigating recomposition, stability, or jank? Start with [`compose-performance`](skills/compose-performance/SKILL.md).
  63. - Comparing Android benchmark configurations or a measured Android default? Start with [`android-benchmark-comparison`](skills/android-benchmark-comparison/SKILL.md).
  64. - Reviewing Flow or coroutine architecture? Start with [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md).
  65. ### Routing
  66. - [`using-chrisbanes-skills`](skills/using-chrisbanes-skills/SKILL.md) — route Kotlin and Jetpack Compose work to focused skills, adding a second only for an independent decision in the same change.
  67. ### Benchmarking
  68. - [`android-benchmark-comparison`](skills/android-benchmark-comparison/SKILL.md) — compare physical Android benchmark configurations with verified coverage, controlled conditions, trace-backed diagnosis of unstable rankings, and bounded conclusions.
  69. ### Jetpack Compose
  70. #### State and side effects
  71. - [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md) — decide state ownership and effect lifecycle for local UI state, screen state holders, Flow collection, callbacks, cleanup, navigation, snackbar, analytics, and focus requests.
  72. #### Performance
  73. - [`compose-performance`](skills/compose-performance/SKILL.md) — diagnose stability, deferred reads, composition contracts, and cross-phase back-writing from concrete runtime evidence.
  74. #### UI API design and layout
  75. - [`compose-component-design`](skills/compose-component-design/SKILL.md) — design caller-placeable Compose APIs whose variable visual regions are caller-provided slots.
  76. - [`compose-animations`](skills/compose-animations/SKILL.md) — choose Compose animation APIs for visibility, value targets, coordinated transitions, and content swaps; align with official quick guide and decision tree.
  77. - [`compose-focus-navigation`](skills/compose-focus-navigation/SKILL.md) — design and test keyboard, TV, D-pad, and focus-first Compose navigation behavior.
  78. #### Testing
  79. - [`compose-ui-testing-patterns`](skills/compose-ui-testing-patterns/SKILL.md) — choose between plain UI tests, semantics assertions, key/focus tests, interaction state tests with MutableInteractionSource, screenshot tests, and integration tests.
  80. ### Kotlin
  81. - [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md) — review coroutine, raw `Thread`, and `Executor` ownership, cancellation, Flow state/event modeling, sharing, replay, and one-shot delivery.
  82. - [`kotlin-control-flow`](skills/kotlin-control-flow/SKILL.md) — write and review Kotlin branching with subject `when`, guard conditions, sealed exhaustiveness, smart casts, nullable branching, and early returns.
  83. - [`kotlin-api-design`](skills/kotlin-api-design/SKILL.md) — choose function owners, semantic domain types, and Kotlin Multiplatform platform boundaries.
  84. ### Writing
  85. - [`changelog`](skills/changelog/SKILL.md) — reconcile Unreleased or specified-release notes with verified history, issue/PR links and contributor credit; preserve curated text and prerelease history across languages and publishing systems.
  86. - [`grounded-writing`](skills/grounded-writing/SKILL.md) — draft or review public developer documentation and other user-owned text using Chris Banes’s writing as the sole style model, including internal report reviews, while preserving evidence, format, review-only authority, and material-edit restraint.
  87. ### Workflows
  88. - [`deliver-issue`](skills/deliver-issue/SKILL.md) — coordinate a named tracker issue and its descendants, or a verified board-controller handoff through solo or justified delegated implementation, a planning-free fast path for small well-understood tickets, CI-supplied full-suite evidence, concurrent external review with verified draft support or earlier PR readiness, reusable scope approval and revision-isolated evidence with preservation checks, independent integrated-candidate review and focused repair review, and PR shepherding, delivering to GitHub or GitLab code hosts; validate applicable live-qualification harness paths offline (isolated production stores for persistent paths, fake runtimes for asynchronous paths), then reconcile failures under bounded approved allowances.
  89. - [`release-kotlin-library`](skills/release-kotlin-library/SKILL.md) — assess readiness, prepare, and verify Kotlin library releases; check the `gradle-maven-publish-plugin` prerequisite, use `changelog` for notes, prepare Metalava API snapshots, and follow repository checks and publication gates.
  90. - [`gradle-run`](skills/gradle-run/SKILL.md) — run every agent-initiated Gradle command through a compact-output wrapper; recover run identity and bounded results by workflow ID with read-only status, and diagnose failures with optional investigation helpers.
  91. - [`subagent-delegation`](skills/subagent-delegation/SKILL.md) — provider-neutral delegation playbook, loaded by name or through a one-line user-level opt-in, with separate Codex and Claude Code model routing, installed-role precedence, parallel briefs, independent audits, and lead acceptance.
  92. - [`implement-with-subagents`](skills/implement-with-subagents/SKILL.md) — default to the lead for a single item or serial graph, apply requested implementation model and effort at preflight, and use workers for every item in a graph with parallel work. Preserve dependency-aware integration, independent final review, same-owner repair, bounded checkout reuse, and read-only orchestration review.
  93. - [`to-plan`](skills/to-plan/SKILL.md) — create risk-scaled implementation plans that prove baselines from green CI, with observable slices and consequential boundary validation before dependent work, plus authenticated integration amendments, tracker-bound markers and explicit consumer handoffs; decide separate PRs by coherent deliverable boundaries.
  94. - [`run-project`](skills/run-project/SKILL.md) — explicitly triage, plan and deliver eligible tickets from a Project board, using Todo / In progress / Done, human-applied role authority, preserved label stops, scheduling and review gates, with a GitHub or GitLab code host; use `deliver-issue` for one named item.
  95. - [`shepherd`](skills/shepherd/SKILL.md) — poll open PRs and MRs directly in the controller, using a read-only helper only for a bounded large failed-log question, or follow a host PR monitor's wake events (such as Claude desktop's Auto-fix) instead of polling, triage review comments (including GitLab discussion reply, resolve and approval reads), and repair CI failures with affected checks and required integrated-candidate validation; start external review during CI while retaining final-head merge gates.
  96. Both delivery skills bundle their internal references for selective installation.
  97. Install `changelog` alongside `release-kotlin-library`.
  98. The delivery skills support Jira Cloud as a tracker for JQL boards, direct
  99. issue delivery and planning markers. Select the code host separately.
  100. See the [Jira binding](references/jira-binding.md) for setup and supported operations.
  101. When `run-project` has no explicit GitHub Project scope, it selects the sole
  102. visible open, non-template Project linked to the verified repository. Supply a
  103. Project URL when discovery is inconclusive.
  104. Delivery workflow migration: `run-github-project` is replaced by `run-project`;
  105. `deliver-tasks` is split into board execution and `deliver-issue`; `deliver-spec`
  106. is folded into their internal delivery procedures. All three old skill names are
  107. removed without aliases. Existing configuration paths and durable records retain
  108. their identities. Local specs use the external `implement` workflow.
  109. Workflows that delegate agents share the [subagent selection and handoff
  110. reference](references/subagent-selection.md). Each workflow retains its own
  111. delegation trigger, authority, and acceptance rules. Each consuming skill also
  112. contains a copy of the reference for standalone installation.
  113. Implementation model and effort requests apply to the agent who edits across
  114. `implement-with-subagents`, `deliver-issue`, and `run-project`, including the
  115. delivery fast path.
  116. `deliver-issue` bundles its own solo-or-delegated implementation procedure;
  117. `implement-with-subagents` chooses solo or delegated ownership once from the
  118. whole task graph.
  119. Planning and drain admission distinguish behavioral prerequisites, shared-file
  120. mutation conflicts and final integration requirements. Plan-proven independent
  121. preparation needs explicit slice ownership, permitted paths/resources, focused
  122. checks and a stop before shared edits; native blockers, accepted-plan ordering,
  123. resource grants and final integrated validation still apply.
  124. Workflows that create worktrees (`deliver-issue`, `implement-with-subagents`
  125. and `run-project`) likewise share the
  126. [checkout lifecycle](references/checkout-lifecycle.md).
  127. Edit `references/subagent-selection.md` or `references/checkout-lifecycle.md`,
  128. then run `npm run references:sync` and include the updated bundled copies in
  129. the change.
  130. Run `npm run references:check` to check that all copies are regular files matching their
  131. sources; it fails on drift without rewriting files.
  132. `.gitattributes` marks every copy as generated so GitHub collapses it in pull
  133. request diffs; a test keeps that list in step with the sync script.
  134. ### Delegation on another machine
  135. Install a revision of `chrisbanes-skills` containing `subagent-delegation`, then
  136. opt in by adding this line to `~/.codex/AGENTS.md` (or `~/.claude/CLAUDE.md`
  137. for Claude Code):
  138. ```markdown
  139. Use `chrisbanes-skills:subagent-delegation` before delegating work or deciding whether to delegate.
  140. ```
  141. Without that line, the skill loads only when invoked by name
  142. (`$chrisbanes-skills:subagent-delegation` in Codex or
  143. `/chrisbanes-skills:subagent-delegation` in Claude Code); its description keeps
  144. it from loading merely because a task involves subagents. No custom Codex or
  145. Claude agent definitions are required, and existing custom agent pins remain
  146. authoritative. Verify available roles, model aliases, effort controls, and
  147. permissions on each machine. `implement-with-subagents` continues to own
  148. implementation orchestration.
  149. ### Migration from pre-cluster skills
  150. This is a breaking taxonomy change. Replace the removed entrypoints as follows:
  151. | Removed skills | Replacement |
  152. |---|---|
  153. | `compose-state-authoring`, `compose-state-hoisting`, `compose-side-effects` | [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md) |
  154. | `compose-recomposition-performance`, `compose-stability-diagnostics`, `compose-state-deferred-reads` | [`compose-performance`](skills/compose-performance/SKILL.md) |
  155. | `compose-modifier-and-layout-style`, `compose-slot-api-pattern` | [`compose-component-design`](skills/compose-component-design/SKILL.md) |
  156. | `kotlin-coroutines-structured-concurrency`, `kotlin-flow-state-event-modeling` | [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md) |
  157. | `kotlin-functions`, `kotlin-types-value-class`, `kotlin-multiplatform-expect-actual` | [`kotlin-api-design`](skills/kotlin-api-design/SKILL.md) |
  158. ## Contributing
  159. Skills live at `skills/<skill-name>/SKILL.md`, flat (no language nesting). The `name:` in the SKILL.md frontmatter must match the directory name.
  160. Frontmatter is validated against [`skills.schema.json`](skills.schema.json), which
  161. tracks the core [Agent Skills specification](https://agentskills.io/specification)
  162. and permits `disable-model-invocation` for Claude Code compatibility.
  163. `name` and `description` are required; portable optional fields are `license`,
  164. `compatibility`, `metadata`, and `allowed-tools`. Explicit-only workflow skills
  165. also mirror that policy in Codex's `agents/openai.yaml`.
  166. ### Releases
  167. Release versions use CalVer: `YYYY.M.D`, `YYYY.M.D.N`, or `YYYY.M.D.NN`, without
  168. zero-padded month or day values. For example, use `2026.6.17` for the first
  169. release of the day and `2026.6.17.1` or `2026.6.17.01` for another release.
  170. Single-digit daily release numbers are normalized to the padded form, so both
  171. inputs produce `2026.6.17.01`.
  172. Keep root `plugin.json`, `.claude-plugin/plugin.json`,
  173. `.codex-plugin/plugin.json`, and new Git release tags on the same version.
  174. Existing zero-padded tags from before this policy map to the non-padded manifest
  175. version, so `2026.06.16` maps to `2026.6.16`. Only bump versions when publishing
  176. an installable release.
  177. To publish a release, run the **Release** workflow from GitHub Actions. Leave the
  178. version input empty to choose today's next unused UTC version, or provide a specific
  179. CalVer value, optionally with a one- or two-digit daily release number. Use the
  180. dry-run option to validate without creating a commit, tag, or GitHub release.
  181. Automatic numbering checks existing tags and releases: the first release uses
  182. `YYYY.M.D`, then subsequent releases increment the highest daily suffix through
  183. `.99`. Gaps are not reused. If all daily numbers are exhausted, the workflow
  184. fails and requires an explicit version or a release on another day.
  185. The same workflow also runs nightly (02:17 UTC). It releases only if `skills/`,
  186. the plugin manifests, `.opencode/`, `.agents/`, `plugin.json`, or `package.json`
  187. changed since the latest tag; otherwise it exits without a release.
  188. Before pushing, lint skills (frontmatter schema + markdown):
  189. ```
  190. npm install
  191. npm run lint
  192. ```
  193. This also runs on CI for all PRs.
  194. For a taxonomy change, also run the durable
  195. [cluster behavior evaluation](tests/cluster-behavior.md). It checks routing,
  196. required references, safeguards, exceptions, and finish gates at the public
  197. agent-facing seam.
  198. Before publishing a release, manually run the advisory evaluations for the
  199. changed skills. Select each affected suite; when shared evaluation machinery
  200. changes, include every suite affected by that change. Preview each suite with
  201. the intended filters and current subject and judge cost assumptions:
  202. ```sh
  203. python3 evals/run.py plan \
  204. --suite <suite> \
  205. --skill <changed-skill> \
  206. --model <model> \
  207. --reasoning <effort> \
  208. --judge-model <judge-model> \
  209. --judge-reasoning <effort> \
  210. --subject-cost-per-call-usd <amount> \
  211. --judge-cost-per-call-usd <amount> \
  212. --json
  213. ```
  214. Repeat `--skill` for each changed skill. Without a `--case` filter, omitting
  215. `--skill` selects all non-calibration cases in the suite. Calibration cases are
  216. excluded unless you pass their explicit IDs with `--case`. Inspect `case_ids` in
  217. the JSON output and confirm the selected cases match the intended scope before
  218. reviewing call counts and estimated cost. The `plan` preview includes only the
  219. counts and cost for first attempts. Execution may automatically retry each
  220. subject and judge call once. Before `--execute`, get explicit cost approval for
  221. up to twice the previewed subject calls, judge calls, and estimated cost, to
  222. cover one retry per call. Execute the matching `run` command manually. Report
  223. invalid or inconclusive results. Before any rerun, preview again, inspect its
  224. `case_ids`, review the calls and cost, and get new approval for the same retry
  225. headroom. Results remain advisory, not release gates. See
  226. [`evals/README.md`](evals/README.md) for suite selection and command options.
  227. ## Evaluating skills
  228. The advisory evaluator tests concrete scenarios modelled on real-world coding
  229. work, with expected outcomes and no-change controls. It compares no-skill,
  230. forced-skill, and automatic-routing runs. **Baseline** and **automatic** use
  231. the cases eligible for automatic activation; **restraint** checks that a skill
  232. does not make an unnecessary change. Scorecards also compare subject-side tokens,
  233. tool calls, completed turns, elapsed time, and total attempted work per
  234. successful outcome. The
  235. table reports the latest available result for each skill and correctness metric.
  236. These scores were produced using
  237. [`gpt-6-luna`](https://developers.openai.com/api/docs/models/gpt-6-luna)
  238. with high reasoning, judged by
  239. [`gpt-6-sol`](https://developers.openai.com/api/docs/models/gpt-6-sol) with
  240. high reasoning. Results are model- and reasoning-specific; other configurations
  241. may perform differently. The human audit queue remains open. These are not merge
  242. or release gates. The subject and judge each run on Codex by default, or on
  243. Claude Code with `--runtime claude` and `--judge-runtime claude`; Claude runs need
  244. an `--output-dir` outside the repository. See
  245. [`evals/README.md`](evals/README.md) for evaluation setup and reproducibility.
  246. Skill-revision compatibility checks compare old and revised instructions within
  247. each model, separately from these benchmark scores. See the
  248. [workflow compatibility record](evals/artifacts/2026-09-06-workflow-model-compatibility.md)
  249. for Astra and 5.6 coverage and its current evidence limits.
  250. | Skill | Baseline | Automatic | Restraint |
  251. | --- | ---: | ---: | ---: |
  252. | [`compose-animations`](skills/compose-animations/SKILL.md) | 75.0% | 100.0% | 100.0% |
  253. | [`compose-component-design`](skills/compose-component-design/SKILL.md) | 86.7% | 100.0% | 100.0% |
  254. | [`compose-focus-navigation`](skills/compose-focus-navigation/SKILL.md) | 33.3% | 100.0% | 100.0% |
  255. | [`compose-performance`](skills/compose-performance/SKILL.md) | 83.3% | 100.0% | 100.0% |
  256. | [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md) | 83.3% | 100.0% | 100.0% |
  257. | [`compose-ui-testing-patterns`](skills/compose-ui-testing-patterns/SKILL.md) | 55.6% | 100.0% | 100.0% |
  258. | [`gradle-run`](skills/gradle-run/SKILL.md) | 41.7% | 100.0% | 100.0% |
  259. | [`kotlin-api-design`](skills/kotlin-api-design/SKILL.md) | 58.3% | 100.0% | 100.0% |
  260. | [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md) | 44.4% | 100.0% | 100.0% |
  261. | [`kotlin-control-flow`](skills/kotlin-control-flow/SKILL.md) | 33.3% | 100.0% | 100.0% |
  262. | [`android-benchmark-comparison`](skills/android-benchmark-comparison/SKILL.md) | 33.3% | 100.0% | 100.0% |
  263. | [`deliver-issue`](skills/deliver-issue/SKILL.md) | — | — | — |
  264. | [`run-project`](skills/run-project/SKILL.md) | — | — | — |
  265. | [`grounded-writing`](skills/grounded-writing/SKILL.md) | 0.0% | 100.0% | 100.0% |
  266. | [`implement-with-subagents`](skills/implement-with-subagents/SKILL.md) | — | — | 100.0% |
  267. | [`release-kotlin-library`](skills/release-kotlin-library/SKILL.md) | 0.0% | 100.0% | 100.0% |
  268. | [`shepherd`](skills/shepherd/SKILL.md) | — | — | 100.0% |
  269. | [`to-plan`](skills/to-plan/SKILL.md) | — | — | 100.0% |
  270. The `android-benchmark-comparison`, `compose-state-and-effects`,
  271. `compose-ui-testing-patterns`, `gradle-run`, `grounded-writing`, `kotlin-api-design`,
  272. `kotlin-concurrency-and-flow`, `kotlin-control-flow`, and
  273. `release-kotlin-library` automatic cells, and the `to-plan` restraint cell,
  274. use later focused evidence. Baseline and efficiency values use the complete
  275. suite. The [improvement result record](evals/artifacts/2026-09-24-gpt6-improvement-results.md),
  276. [targeted probe record](evals/artifacts/2026-09-24-gpt6-targeted-100-probes.md),
  277. and [inline repair record](evals/artifacts/2026-09-24-gpt6-inline-repair-results.md)
  278. give provenance and remaining failures.
  279. ### Skill efficiency
  280. Values are per-run medians, baseline → automatic, followed by the automatic
  281. percentage change. These subject-only measurements use the latest complete,
  282. same-run evidence available for each suite and include failed runs and negative
  283. controls. Baseline-to-automatic efficiency comparisons use only cases eligible
  284. for automatic activation. Multi-skill scenarios contribute to every targeted
  285. skill row. A turn is one completed top-level turn (a Codex `turn.completed` or a
  286. Claude Code `result`); time remains environment-sensitive.
  287. Run provenance and local scorecard paths are in the
  288. [GPT-6 improvement result record](evals/artifacts/2026-09-24-gpt6-improvement-results.md).
  289. | Skill | Tokens / run | Tool calls / run | Turns / run | Time / run |
  290. | --- | ---: | ---: | ---: | ---: |
  291. | [`compose-animations`](skills/compose-animations/SKILL.md) | 58.6k → 85.4k (+46%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 28.2s → 34.8s (+23%) |
  292. | [`compose-component-design`](skills/compose-component-design/SKILL.md) | 48.7k → 73.4k (+51%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 24.4s → 29.5s (+21%) |
  293. | [`compose-focus-navigation`](skills/compose-focus-navigation/SKILL.md) | 59.1k → 84.9k (+44%) | 5 → 5 (+0%) | 1 → 1 (+0%) | 28.9s → 36.6s (+27%) |
  294. | [`compose-performance`](skills/compose-performance/SKILL.md) | 49.1k → 85.0k (+73%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 26.4s → 31.8s (+21%) |
  295. | [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md) | 60.1k → 89.0k (+48%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 28.2s → 37.8s (+34%) |
  296. | [`compose-ui-testing-patterns`](skills/compose-ui-testing-patterns/SKILL.md) | 60.2k → 84.2k (+40%) | 5.5 → 5 (-9%) | 1 → 1 (+0%) | 27.1s → 26.8s (-1%) |
  297. | [`gradle-run`](skills/gradle-run/SKILL.md) | 59.8k → 104.3k (+75%) | 4 → 7 (+75%) | 1 → 1 (+0%) | 27.6s → 45.3s (+64%) |
  298. | [`kotlin-api-design`](skills/kotlin-api-design/SKILL.md) | 60.1k → 83.9k (+40%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 32.8s → 38.8s (+18%) |
  299. | [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md) | 60.0k → 74.0k (+23%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 26.5s → 28.0s (+6%) |
  300. | [`kotlin-control-flow`](skills/kotlin-control-flow/SKILL.md) | 60.9k → 83.9k (+38%) | 5 → 5 (+0%) | 1 → 1 (+0%) | 26.2s → 38.5s (+47%) |
  301. | [`android-benchmark-comparison`](skills/android-benchmark-comparison/SKILL.md) | 46.9k → 53.8k (+15%) | 3 → 3 (+0%) | 1 → 1 (+0%) | 23.2s → 35.0s (+51%) |
  302. | [`grounded-writing`](skills/grounded-writing/SKILL.md) | 35.6k → 56.9k (+60%) | 2 → 4 (+100%) | 1 → 1 (+0%) | 23.2s → 22.1s (-5%) |
  303. | [`release-kotlin-library`](skills/release-kotlin-library/SKILL.md) | 70.2k → 88.3k (+26%) | 4 → 5 (+25%) | 1 → 1 (+0%) | 42.1s → 43.9s (+4%) |
  304. ### Claude Code
  305. Claude Code: `claude-haiku-5-5` with high effort, judged by `claude-sonnet-5-5` with high
  306. effort on Claude Code. The metrics are defined as above, and `—` marks a cell with no
  307. recorded result. Run provenance and scorecards are in the
  308. [Claude Code result record](evals/artifacts/2026-10-19-claude-haiku-5-5-sonnet-judge-results.md).
  309. | Skill | Baseline | Automatic | Restraint |
  310. | --- | ---: | ---: | ---: |
  311. | [`compose-animations`](skills/compose-animations/SKILL.md) | 91.7% | 100.0% | 100.0% |
  312. | [`compose-component-design`](skills/compose-component-design/SKILL.md) | 86.7% | 100.0% | 100.0% |
  313. | [`compose-focus-navigation`](skills/compose-focus-navigation/SKILL.md) | 100.0% | 100.0% | 100.0% |
  314. | [`compose-performance`](skills/compose-performance/SKILL.md) | 91.7% | 100.0% | 100.0% |
  315. | [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md) | 79.2% | 95.8% | 100.0% |
  316. | [`compose-ui-testing-patterns`](skills/compose-ui-testing-patterns/SKILL.md) | 88.9% | 100.0% | 100.0% |
  317. | [`gradle-run`](skills/gradle-run/SKILL.md) | 16.7% | 83.3% | 100.0% |
  318. | [`kotlin-api-design`](skills/kotlin-api-design/SKILL.md) | 91.7% | 91.7% | 100.0% |
  319. | [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md) | 72.2% | 100.0% | 100.0% |
  320. | [`kotlin-control-flow`](skills/kotlin-control-flow/SKILL.md) | 61.1% | 94.4% | 100.0% |
  321. | [`android-benchmark-comparison`](skills/android-benchmark-comparison/SKILL.md) | 33.3% | 100.0% | 100.0% |
  322. | [`deliver-issue`](skills/deliver-issue/SKILL.md) | — | — | — |
  323. | [`run-project`](skills/run-project/SKILL.md) | — | — | — |
  324. | [`grounded-writing`](skills/grounded-writing/SKILL.md) | 0.0% | 100.0% | 100.0% |
  325. | [`implement-with-subagents`](skills/implement-with-subagents/SKILL.md) | — | — | 100.0% |
  326. | [`release-kotlin-library`](skills/release-kotlin-library/SKILL.md) | 16.7% | 100.0% | 100.0% |
  327. | [`shepherd`](skills/shepherd/SKILL.md) | — | — | 100.0% |
  328. | [`to-plan`](skills/to-plan/SKILL.md) | 16.7% | 100.0% | 100.0% |
  329. Per-run medians, baseline → automatic, for the same configuration:
  330. | Skill | Tokens / run | Tool calls / run | Turns / run | Time / run |
  331. | --- | ---: | ---: | ---: | ---: |
  332. | [`compose-animations`](skills/compose-animations/SKILL.md) | 51.8k → 85.9k (+66%) | 6 → 8 (+33%) | 1 → 1 (+0%) | 47.7s → 46.4s (-3%) |
  333. | [`compose-component-design`](skills/compose-component-design/SKILL.md) | 52.2k → 72.8k (+40%) | 6 → 6 (+0%) | 1 → 1 (+0%) | 30.2s → 35.2s (+16%) |
  334. | [`compose-focus-navigation`](skills/compose-focus-navigation/SKILL.md) | 62.2k → 88.9k (+43%) | 7.5 → 8 (+7%) | 1 → 1 (+0%) | 44.6s → 56.9s (+27%) |
  335. | [`compose-performance`](skills/compose-performance/SKILL.md) | 94.3k → 103.8k (+10%) | 8 → 9 (+12%) | 1 → 1 (+0%) | 45.0s → 51.3s (+14%) |
  336. | [`compose-state-and-effects`](skills/compose-state-and-effects/SKILL.md) | 72.4k → 118.4k (+64%) | 7 → 9 (+29%) | 1 → 1 (+0%) | 39.4s → 51.3s (+30%) |
  337. | [`compose-ui-testing-patterns`](skills/compose-ui-testing-patterns/SKILL.md) | 87.0k → 88.3k (+2%) | 10 → 8 (-20%) | 1 → 1 (+0%) | 39.3s → 44.7s (+13%) |
  338. | [`gradle-run`](skills/gradle-run/SKILL.md) | 75.6k → 105.6k (+40%) | 6 → 5 (-17%) | 1 → 1 (+0%) | 36.3s → 29.9s (-18%) |
  339. | [`kotlin-api-design`](skills/kotlin-api-design/SKILL.md) | 96.7k → 101.8k (+5%) | 8 → 9 (+12%) | 1 → 1 (+0%) | 45.4s → 53.5s (+18%) |
  340. | [`kotlin-concurrency-and-flow`](skills/kotlin-concurrency-and-flow/SKILL.md) | 111.2k → 145.2k (+31%) | 9.5 → 10 (+5%) | 1 → 1 (+0%) | 47.1s → 50.4s (+7%) |
  341. | [`kotlin-control-flow`](skills/kotlin-control-flow/SKILL.md) | 119.5k → 138.1k (+16%) | 10 → 10 (+0%) | 1 → 1 (+0%) | 41.3s → 44.1s (+7%) |
  342. | [`android-benchmark-comparison`](skills/android-benchmark-comparison/SKILL.md) | 37.7k → 66.3k (+76%) | 3 → 3 (+0%) | 1 → 1 (+0%) | 22.0s → 22.6s (+3%) |
  343. | [`grounded-writing`](skills/grounded-writing/SKILL.md) | 46.6k → 73.6k (+58%) | 3 → 5 (+67%) | 1 → 1 (+0%) | 19.2s → 28.3s (+47%) |
  344. | [`release-kotlin-library`](skills/release-kotlin-library/SKILL.md) | 55.5k → 60.7k (+9%) | 5 → 3 (-40%) | 1 → 1 (+0%) | 43.6s → 54.9s (+26%) |
  345. | [`to-plan`](skills/to-plan/SKILL.md) | 22.9k → 53.4k (+133%) | 2 → 4 (+100%) | 1 → 1 (+0%) | 9.4s → 21.8s (+132%) |
  346. ## License
  347. [Apache 2.0](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.