Releases: LightDevCoder/skills
Release list
v0.1.6 — kb-init & kanban-worker
What's new in this release
- Renamed
light-kanban-worker→kanban-workerso the skill is shorter and easier to invoke. - Greatly optimized
kb-init's internal logic: decision provenance, open-decision surfacing, depth-before-settlement, readiness checks, human-navigation design, a research contract, connection setup/validation, and backup/recovery semantics. - Skills no longer carry version suffixes.
kb-init
Knowledge-base initialization skill. It runs a knowledge-base-specific interview, researches how the chosen base can actually be operated, produces an implementation SPEC, and only starts implementation, validation, and handoff after you explicitly approve. When invoked with little context, it starts the interview automatically.
kanban-worker
Board-work processing skill. Each scheduled wake handles one Light-Kanban task: continues owned in-progress work and review feedback first, claims new work, validates the workspace, executes, and returns the result for human confirmation. A manual entry point is supported.
Verification
Fresh whole-collection and per-Skill installs verified for kanban-worker and kb-init; collection-quality CI passes.
中文
v0.1.5 — light-kanban-worker scheduling & identity contract
light-kanban-worker v0.1.5 — maintenance release
What changed
- same-agent scheduled runs must not overlap — at most one invocation per agent id may be active; a wake that fires while the previous run is still active must skip. Different agent ids may still run concurrently.
- scheduler concurrency boundary documented — atomic claim is not a concurrency lock for multiple invocations using the same agent identity; concurrency control stays with the scheduler / agent runtime (
max concurrent runs = 1or an equivalent skip-while-active setting). The worker adds no lock process, heartbeat, or lease service. - first-registration avatar requirements clarified — first registration needs ID + name + avatar (local images via
POST /api/avatars); later runs reuse the stored identity. A new agent id without name/avatar reports identity configuration missing and mutates nothing. - new overlap / identity tests — contract suite extended (100 assertions), two new adversarial negative fixtures, behavior scenarios G (same-agent concurrent wake) and H (fresh identity without avatar); scenarios A–F unchanged.
- release-evidence workflow clarified — the receipt now separates the pre-release gate (
READY FOR RELEASE) from post-release verification (published tag, fresh install, discovery, CI), recorded on main.
Evidence
- Admission evidence: https://github.com/LightDevCoder/skills/tree/main/docs/evidence/admissions/light-kanban-worker
- Agent-skill review PASS: https://github.com/LightDevCoder/skills/blob/main/docs/evidence/releases/v0.1.5/AGENT_SKILL_REVIEW.md
- Published-tag installation verification: https://github.com/LightDevCoder/skills/blob/main/docs/evidence/releases/v0.1.5/INSTALLATION_VERIFICATION.md
- Post-release receipt (on main): https://github.com/LightDevCoder/skills/blob/main/docs/evidence/releases/v0.1.5/RELEASE_RECEIPT.md
- CI (
collection-quality): PASS, run31985455493on commita56aa9d
Verified install
npx skills add LightDevCoder/skills#v0.1.5 --skill light-kanban-worker --yes --copy --agent '*'
(whole-collection and latest forms also verified; Skills CLI 1.5.22; installed copy byte-identical to this tag)
中文
v0.1.4 — light-kanban-worker: scheduled Light-Kanban worker
What's new
- New first-party
light-kanban-worker(model-invoked, manual entry supported): each scheduled agent run processes at most one Light-Kanban task. Owned in-progress work andreviewFeedback(Request Changes) are checked before any new claim; new work is claimed atomically from the FIFO To Do queue with bounded conflict retry; an inaccessible workspace becomesblockwith a meaningful reason;completereturns the work to human confirmation. The worker never archives, accepts, deletes, recycles, or unblocks tasks, and never loops or runs a daemon. - Scheduled Light-Kanban worker workflow: install the Skill, create work on the board, schedule an agent with a one-line prompt ("Use light-kanban-worker to process at most one Light-Kanban task."), and review results in Awaiting Confirmation.
- Review-feedback resume: Request Changes rework is picked up by the same agent on its next wake — no new task is created.
- Atomic multi-agent claim support: concurrent workers race safely; exactly one claim wins, the loser retries bounded and exits.
- Cross-platform scanner fix: ask-light's readable-path check no longer assumes Windows separators (collection-quality CI green on ubuntu-latest again).
- Catalog / installation version synchronization: v0.1.4 is the current stable release; v0.1.3 and earlier remain historical records.
Verification
- Admission:
review-loop agent-skillPASS(full independence; see admission evidence) - Behavioral scenarios A–F against a real Light-Kanban server:
PASS - Collection-quality CI:
PASSon commita9cc8aa - Fresh installation verification:
PASSwith Skills CLI1.5.22for bothlatestand#v0.1.4forms (whole collection and per-Skill)
Compatible with Light-Kanban v1.0.4+; v1.0.5 is the recommended integration version.
中文
v0.1.3 — cross-platform test toolchain
Test toolchain migration release.
- 21 PowerShell test files replaced by 18 Python suites (assertions preserved)
- CI moved to ubuntu-latest (bash + python)
- ask-light scanner behavior runs the real ask-light.ps1 via pwsh when available
- No governance wording changes; docs updated for cross-platform references
- Evidence: docs/evidence/releases/v0.1.3/
Follow-up fixes included in the tag:
- collection-discovery suite wired into 'unittest discover' (was silently absent from CI)
- directory-link semantics and manual-fallback check corrected in the port
- zero-assertion gate enforced; changelog entries added
中文
v0.1.2 — Personal Skills Collection
English
v0.1.2 is the published first-party Personal Skills Collection release.
- The released collection contains exactly seven first-party packages: the five in v0.1.1 (
ask-light,learn-anything,manuscript-ops,project-init,review-loop) plusrecapandlanguage-learning. recapandlanguage-learningwere admitted through the prompt-only fast track with independent EvaluatorPASS.- Install guidance uses generic
latestcommands (--agent '*', no version), with pinned#v0.1.2retained for reproducible installs. - Fresh whole-collection and per-Skill installs verified
PASSwith CLI1.5.22for both the genericlatestand pinned#v0.1.2forms; see the installation verification. collection-qualityCI passes on the release and publish commits (run31362999381on the tag commit).- Independent
review-loop agent-skillacceptance remainsBLOCKEDfor the original five;recapandlanguage-learningcarry independent fast-trackPASS.
Release evidence: docs/evidence/releases/v0.1.2/.
中文
v0.1.2 是已发布的第一方 Personal Skills Collection 版本。
- 发布集合恰好包含七个第一方包:v0.1.1 的五个包(
ask-light、learn-anything、manuscript-ops、project-init、review-loop)加上recap与language-learning。 recap与language-learning经纯提示型快速通道准入,带有独立 EvaluatorPASS。- 安装说明使用通用
latest命令(--agent '*'、不带版本号),并保留 pinned#v0.1.2用于可复现安装。 - 使用 CLI
1.5.22对通用latest与 pinned#v0.1.2两种形式均验证 fresh 整仓与单 Skill 安装为PASS;见安装验证。 collection-qualityCI 在 release commit 与 publish commit 上均通过(tag commit 上 run31362999381)。- 原有五个包的独立
review-loop agent-skillacceptance 仍为BLOCKED;recap与language-learning带有独立 fast-trackPASS。
发布证据见 docs/evidence/releases/v0.1.2/。
中文
v0.1.1 — Personal Skills Collection
English
v0.1.1 publishes the public first-party Personal Skills Collection at
release commit c50f1ef403a5f0bfe02e75d1aeff2c237556db63.
- The flat collection contains exactly five first-party packages:
ask-light,learn-anything,manuscript-ops,project-init, andreview-loop. - The release includes the layered
1600 × 480SVG/PNG header, bilingual user guides, workflow recipes, Quick Start, metadata policy repair, and GitHub Actions quality checks. - Skills CLI
1.5.20fresh-install verification passed for the whole collection and a singlereview-looppackage; source checkouts were absent from both destinations. - Release-commit Actions run
30189210521passed. - Independent
review-loop agent-skillacceptance remainsBLOCKEDuntil a separate evaluator record exists.
Release evidence: docs/evidence/releases/v0.1.1/.
中文
v0.1.1 发布公开的第一方 Personal Skills Collection,release commit 为
c50f1ef403a5f0bfe02e75d1aeff2c237556db63。
- 平面集合恰好包含五个第一方包:
ask-light、learn-anything、manuscript-ops、project-init和review-loop。 - 本版本包含
1600 × 480叠层 SVG/PNG 头图、双语用户指南、workflow recipes、Quick Start、metadata policy 修复和 GitHub Actions quality checks。 - Skills CLI
1.5.20已通过整仓和单个review-loop的 fresh-install 验证;两个 destination 都不存在 source checkout。 - Release-commit Actions run
30189210521已通过。 - 在取得独立 evaluator 记录前,独立
review-loop agent-skillacceptance 仍为BLOCKED。
发布证据见 docs/evidence/releases/v0.1.1/。
中文
v0.1.0
Stable first-party Personal Skills Collection release. Includes ask-light, learn-anything, manuscript-ops, project-init, and review-loop, with the published header asset and documented installation commands.