Repository navigation
Releases
prompt-architect v0.3.0 (experimental prerelease)
Prompt Architect builds a task method, reads the target platform's current official guidance, and writes one ready-to-paste prompt with source notes. This release changes Step 2 of the method.
Experimental v0.3.0 · Ansel · MIT. Maintained by Ansel, with no promised response time.
What changed:
- Step 2 is now "Break down the task". The prompt states the task's sub-tasks, the decisions they call for and what each decision rests on, instead of a script of reasoning steps. Whether the prompt should ask the model to reason step by step or check its own work is now decided in Step 5, from the vendor's pages for the target model.
SOURCES.mdnow credits the task method as an arrangement by Trigpoint (2026) whose components follow the vendors' own guidance.- Validation check 1 now asks for a real breakdown of the task.
Download the attached prompt-architect-v0.3.0.zip, unzip it, and copy the entire prompt-architect/ folder to your host's supported skill directory. Keep its references and MIT license. SHA256SUMS.txt records the ZIP checksum. GitHub's automatically generated source archives contain the whole collection; use skills/prompt-architect/ if you choose those instead.
Tested on Claude Code on macOS, with the skill path given to fresh sub-agents rather than installed: practice requests R1 to R6 and three requests that name the model, each run once and scored by the coordinating agent, not blind. Eight passed; one failed on a usage note that came from the test host's own stored notes. Installation was last tested for 0.2.0, on Codex. The generated prompts have not been evaluated on their target platforms with real inputs. See the validation record.
本版為 experimental v0.3.0:方法第 2 步改為「拆解任務」,只寫任務本身的組成;要不要求模型逐步推理或自我檢查,改在第 5 步依目標平台的官方頁決定。
Questions and reproducible problems: Issues. Sensitive security reports: private vulnerability reporting.
classify-ai-incident v0.1.0 (experimental prerelease)
0.1.0, tag classify-ai-incident-v0.1.0, 2026-11-02
First public release, experimental. See docs/classify-ai-incident/VALIDATION.md for what was and was not tested.
Method article: Classify an AI incident before you fix it.
Download the attached classify-ai-incident-v0.1.0.zip, unzip it, and copy the entire classify-ai-incident/ folder to your host's supported skill directory. SHA256SUMS.txt records the ZIP checksum. GitHub's automatically generated source archives contain the whole collection; use skills/classify-ai-incident/ if you choose those instead.
calibrate-ai-checks v0.1.0 (experimental prerelease)
0.1.0, tag calibrate-ai-checks-v0.1.0, 2026-11-02
First public release, experimental. See docs/calibrate-ai-checks/VALIDATION.md for what was and was not tested.
Method article: Calibrate how often you check AI output.
Download the attached calibrate-ai-checks-v0.1.0.zip, unzip it, and copy the entire calibrate-ai-checks/ folder to your host's supported skill directory. SHA256SUMS.txt records the ZIP checksum. GitHub's automatically generated source archives contain the whole collection; use skills/calibrate-ai-checks/ if you choose those instead.
allocate-ai-work v0.1.0 (experimental prerelease)
0.1.0, tag allocate-ai-work-v0.1.0, 2026-11-02
First public release, experimental. See docs/allocate-ai-work/VALIDATION.md for what was and was not tested.
Method article: Allocate a workflow before you hand it to AI.
Download the attached allocate-ai-work-v0.1.0.zip, unzip it, and copy the entire allocate-ai-work/ folder to your host's supported skill directory. SHA256SUMS.txt records the ZIP checksum. GitHub's automatically generated source archives contain the whole collection; use skills/allocate-ai-work/ if you choose those instead.
prompt-architect v0.2.0 (experimental prerelease)
Prompt Architect builds a task method, reads the target platform's current official guidance, and writes one ready-to-paste prompt with source notes. This release changes how it finds that guidance.
Experimental v0.2.0 · Ansel · MIT. Maintained by Ansel, with no promised response time.
What changed:
- The source list is split into one file per vendor. The skill opens only the target vendor's file, which pairs each point it checks with the page to read first. It still ships no vendor conventions and reads the vendor's pages each time.
- The Claude "Adaptive thinking" entry now uses the page's current title and address, "Steering thinking". All 28 source links were rechecked on 2026-11-01.
- The skill README and the first-use tutorial link to the method article.
Download the attached prompt-architect-v0.2.0.zip, unzip it, and copy the entire prompt-architect/ folder to your host's supported skill directory. Keep its references and MIT license. SHA256SUMS.txt records the ZIP checksum. GitHub's automatically generated source archives contain the whole collection; use skills/prompt-architect/ if you choose those instead.
Tested on Codex desktop on macOS with the skill installed: practice requests R1 to R6 and three requests that name the model, each run once and scored by the coordinating agent, not blind. When a request names only a product (for example "Claude"), this version answers "model unknown" and uses only guidance that does not depend on the model; name the model to get model-specific guidance. The generated prompts have not been evaluated on their target platforms with real inputs. See the validation record.
本版為 experimental v0.2.0:來源清單按廠商拆檔,skill 只讀目標廠商的那一份,仍不內附任何廠商規則。只給產品名時不假定型號;指定型號才會套用型號專屬指引。
Questions and reproducible problems: Issues. Sensitive security reports: private vulnerability reporting.
concept-relations v0.1.0
0.1.0, tag concept-relations-v0.1.0, 2026-10-28
First public release, experimental. Vocabulary v0.1. See docs/concept-relations/VALIDATION.md for what was and was not tested.
prompt-architect v0.1.0 (experimental prerelease)
Prompt Architect is now the fourth independently installable skill in this collection. It builds a task method, reads the target platform's current official guidance, and writes one ready-to-paste prompt with source notes.
Experimental v0.1.0 · Ansel · MIT. Maintained by Ansel, with no promised response time.
Download the attached prompt-architect-v0.1.0.zip, unzip it, and copy the entire prompt-architect/ folder to your host's supported skill directory. Keep its references and MIT license. SHA256SUMS.txt records the ZIP checksum. GitHub's automatically generated source archives contain the whole collection; use skills/prompt-architect/ if you choose those instead.
The instruction body and description text are unchanged from the revised candidate tested on Codex. The description now uses valid quoted YAML. This release finishes the documentation, includes the completed follow-up results, and supplies a bilingual first-use tutorial. All 28 source URLs were reachable on 2026-10-26, including one that required a different reader after HTTP 403.
Codex installation and discovery were tested on macOS. Claude Code was tested by explicit file path, not native installation. The generated prompts have not been evaluated on their target platforms with real inputs. The tutorial's meeting notes and expected results are synthetic teaching material, not an execution transcript. See the validation record for the single-run, non-blind test limits.
本版為 experimental v0.1.0,與中英最短教學同步推出。專屬 Trigpoint 網站文章仍待準備;筆記本頁的 NotebookLM v2.5 是另一個入口,不是本 skill 的版本號。
Questions and reproducible problems: Issues. Sensitive security reports: private vulnerability reporting.
red-team-gate v0.1.0 (experimental prerelease)
red-team-gate
0.1.0, tag red-team-gate-v0.1.0, 2026-10-24
First public release, experimental. Tested on Codex on macOS; see docs/red-team-gate/VALIDATION.md for what was and was not tested.
price-watch v0.1.0 (experimental prerelease)
The content is the same as anselbuilds/safari-price-watch v0.1.0, except for the reporting links and the relocation adjustments listed below.
price-watch
0.1.0, tag price-watch-v0.1.0, 2026-10-24
Moved from anselbuilds/safari-price-watch, release v0.1.0 (commit 408477fe). The installable name price-watch is unchanged. The content is the same as that release, except that the links for reporting issues and security problems now point to this repository, because the old repository is archived.
Relocated files:
.github/ISSUE_TEMPLATE/bug_report.yml->docs/price-watch/.github/ISSUE_TEMPLATE/bug_report.yml..gitignore->docs/price-watch/.gitignore.CHANGELOG.md->docs/price-watch/CHANGELOG.md.README.zh-Hant.md->docs/price-watch/README.zh-Hant.md.SECURITY.md->docs/price-watch/SECURITY.md.VALIDATION.md->docs/price-watch/VALIDATION.md.tests/test_watchlist.py->tests/price-watch/test_watchlist.py.README.md->docs/price-watch/README.md.
The original skills/price-watch/README.md is retained, with only its two explicitly authorized reporting URLs changed; all other bytes match the original. Only relocation paths/relative links and the explicitly listed reporting URLs differ.
Content/path changes:
docs/price-watch/README.zh-Hant.md:](README.md)->](https://github.com/anselbuilds/safari-price-watch/blob/408477fed8ee2b00af4759da69b4c7b90b00d2d2/README.md)(1 occurrence(s)).docs/price-watch/README.zh-Hant.md:](LICENSE)->](../../LICENSE)(1 occurrence(s)).docs/price-watch/README.zh-Hant.md:](skills/price-watch/references/registry.md)->](../../skills/price-watch/references/registry.md)(1 occurrence(s)).tests/price-watch/test_watchlist.py:Path(__file__).resolve().parents[1]->Path(__file__).resolve().parents[2](1 occurrence(s)).docs/price-watch/VALIDATION.md:discover -s tests -v->discover -s tests/price-watch -v(1 occurrence(s)).docs/price-watch/README.zh-Hant.md:discover -s tests -v->discover -s tests/price-watch -v(1 occurrence(s)).docs/price-watch/README.md:](LICENSE)->](../../LICENSE)(1 occurrence(s)).docs/price-watch/README.md:](skills/price-watch/references/registry.md)->](../../skills/price-watch/references/registry.md)(1 occurrence(s)).docs/price-watch/README.md:](skills/price-watch/README.md)->](../../skills/price-watch/README.md)(1 occurrence(s)).docs/price-watch/README.md:discover -s tests -v->discover -s tests/price-watch -v(1 occurrence(s)).docs/price-watch/SECURITY.md:https://github.com/anselbuilds/safari-price-watch/issues->https://github.com/anselbuilds/agent-skills/issues(1 occurrence(s)).docs/price-watch/SECURITY.md:https://github.com/anselbuilds/safari-price-watch/security/advisories/new->https://github.com/anselbuilds/agent-skills/security/advisories/new(1 occurrence(s)).skills/price-watch/README.md:https://github.com/anselbuilds/safari-price-watch/issues->https://github.com/anselbuilds/agent-skills/issues(1 occurrence(s)).skills/price-watch/README.md:https://github.com/anselbuilds/safari-price-watch/security/advisories/new->https://github.com/anselbuilds/agent-skills/security/advisories/new(1 occurrence(s)).
inspect-github-agent-resources v0.1.0 (trial prerelease)
The content is the same as anselbuilds/inspect-github-agent-resources v0.1.0, except for the reporting links and the relocation adjustments listed below.
inspect-github-agent-resources
0.1.0, tag inspect-github-agent-resources-v0.1.0, 2026-10-24
Moved from anselbuilds/inspect-github-agent-resources, release v0.1.0 (commit 2d43edf0). The content is the same as that release, except that the links for reporting issues and security problems now point to this repository, because the old repository is archived.
Relocated files:
.github/ISSUE_TEMPLATE/bug_report.yml->docs/inspect-github-agent-resources/.github/ISSUE_TEMPLATE/bug_report.yml..gitignore->docs/inspect-github-agent-resources/.gitignore.CHANGELOG.md->docs/inspect-github-agent-resources/CHANGELOG.md.SECURITY.md->docs/inspect-github-agent-resources/SECURITY.md.VALIDATION.md->docs/inspect-github-agent-resources/VALIDATION.md.README.md->docs/inspect-github-agent-resources/README.md.
The original skills/inspect-github-agent-resources/README.md is retained byte-for-byte. Only relocation paths/relative links and the explicitly listed reporting URLs differ.
Content/path changes:
docs/inspect-github-agent-resources/README.md:](LICENSE)->](../../LICENSE)(2 occurrence(s)).docs/inspect-github-agent-resources/README.md:](skills/inspect-github-agent-resources/SKILL.md)->](../../skills/inspect-github-agent-resources/SKILL.md)(1 occurrence(s)).docs/inspect-github-agent-resources/README.md:](skills/inspect-github-agent-resources/examples/review-example.md)->](../../skills/inspect-github-agent-resources/examples/review-example.md)(2 occurrence(s)).docs/inspect-github-agent-resources/README.md:](skills/inspect-github-agent-resources/references/scanner.md)->](../../skills/inspect-github-agent-resources/references/scanner.md)(1 occurrence(s)).docs/inspect-github-agent-resources/README.md:](skills/inspect-github-agent-resources/README.md)->](../../skills/inspect-github-agent-resources/README.md)(1 occurrence(s)).docs/inspect-github-agent-resources/SECURITY.md:https://github.com/anselbuilds/inspect-github-agent-resources/issues->https://github.com/anselbuilds/agent-skills/issues(1 occurrence(s)).docs/inspect-github-agent-resources/SECURITY.md:https://github.com/anselbuilds/inspect-github-agent-resources/security/advisories/new->https://github.com/anselbuilds/agent-skills/security/advisories/new(1 occurrence(s)).skills/inspect-github-agent-resources/SECURITY.md:https://github.com/anselbuilds/inspect-github-agent-resources/issues->https://github.com/anselbuilds/agent-skills/issues(1 occurrence(s)).skills/inspect-github-agent-resources/SECURITY.md:https://github.com/anselbuilds/inspect-github-agent-resources/security/advisories/new->https://github.com/anselbuilds/agent-skills/security/advisories/new(1 occurrence(s)).