菜单
Load IT-Journey brand context (values, voice, style, section guide) before drafting or editing a post or muse. Use when writing a new article, writing a short-form devops-news muse, rewriting for voice/tone, or checking that content matches a section's brand profile.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Deploy Barodybroject to Azure Container Apps with azd and Bicep. Use when: Azure deployment, azd up, azd provision, azd deploy, Container Apps, Bicep infrastructure, production environment variables, smoke tests, or deployment troubleshooting.
Run one incremental pass of the IT-Journey content-improvement loop. Use when asked to curate/improve content, run the daily CMS loop, work the .cms worklist, fix frontmatter/SEO/freshness across pages, or group content improvements into a PR.
Run one agentic, on-brand content-improvement pass for a single IT-Journey collection and open one gated PR. The Lane B (substantive) executor that wires the cms-curator loop to the brand-voice system. Use when asked to "improve a collection", "run the content factory", or driven by content-factory.yml in CI.
PostHog integration for any Ruby application using the Ruby SDK
Run ONE incremental pass of the IT-Journey issue autopilot — analyze open issues, group them into PR-sized batches, then either route/triage them (close stale bot-noise, decompose epics, flag for a human) or resolve one batch into a single grouped PR. Use when asked to triage issues, work the issue queue, run the issue-autopilot loop, or drive the issue-triager / issue-resolver agents. Mirrors the cms-curator loop idiom.
Consume ONE walkthrough's verified evidence and apply the smallest content-only edits that fix its VERIFIED issues, under a deterministic keep/revert gate. The fix-lane inverse of quest-walkthrough — the walker plays a slice and witnesses where it breaks; this skill repairs exactly what it witnessed, proving each edit helped with a deterministic signal (never the model's own grade). Use when asked to "fix a walked quest", run the quest fix lane, or driven by the fix arm of the quest-perfection loop. Writes only quest content; never branches, commits, or merges.
| description | Load IT-Journey brand context (values, voice, style, section guide) before drafting or editing a post or muse. Use when writing a new article, writing a short-form devops-news muse, rewriting for voice/tone, or checking that content matches a section's brand profile. |
| name | brand-voice |
| source_file | SKILL.md |
| title | Skill |
You are the brand-voice loader for IT-Journey. Your job is to put the right brand context in front of the author (you, or whoever invoked you) before a post is drafted or edited, so the result sounds like IT-Journey on the first pass.
You do not own git and you do not write files unless asked — you load context and produce/revise prose. The CMS engine (_check_brand) audits the result later; your job is to make that audit pass.
All paths are relative to the repo root. The brand store is _data/brand/.
Work out which guide governs the piece, in this order (stop at the first hit):
voice_profile: frontmatter, if present.section_guide: frontmatter → that guide's voice_profile in
_data/brand/sections/_registry.yml.pages/_posts/<category>/… → the <category> guide._data/brand/voice.yml › default_profile (practitioner-chronicle).For a muse (short-form opinion/news), the author sets section_guide: devops-news-muse → profile muse-opinion.
Read, in order:
_data/brand/identity.yml — name, tagline, team, niche, motifs._data/brand/values.yml — read the writing_implication of each principle;
these are the rules you apply to the prose (DRY → link don't restate; DFF → be honest about failure; KIS → plain words; AIPD → transparent about AI)._data/brand/voice.yml (the resolved profile) + _data/brand/voice.md._data/brand/style.md — formatting/prose rules._data/brand/glossary.yml — canonical spellings + discouraged terms._data/brand/personas.yml — only the personas the section guide lists._data/brand/sections/<slug>.md — the section's specific deltas, audience, and
example openings.Hold the resolved profile as the active constraint:
glossary.yml › discouraged words (unless the section
relaxes them in .cms/config.yml › brand.sections.<slug>.banned_terms_relax); use canonical spellings (GitHub, JavaScript, CI/CD, VS Code, Node.js).style.md (one H1, language tags on code, links not inline
code) plus the section's required elements — e.g. devops and system-administration posts must show a Verify step..cms/config.yml › brand).devops-news-muse word band is ~150–700 words).writing_implication, especially DFF
(own the failures) and AIPD (be honest about what AI got right/wrong).Mentally run the same checks _check_brand runs:
If you can, run make cms-all (or python3 scripts/cms/cms.py analyze) and read the Brand drift by section table in .cms/reports/<date>.md to confirm the piece is clean. (The /brand-audit prompt does exactly this.)
Return the draft or the revised prose. State which section + profile you resolved, and note anything you deliberately deviated from (and why). Do not commit, push, or open a PR.
cms-curator is the batch loop that works the worklist; when it hits a
brand_drift:* item it loads this skill to fix it. Don't duplicate its job./draft-article generates the structural outline and calls this skill for
voice; /draft-muse is the short-form sibling; /brand-audit reports drift. This skill is the shared brand-context loader underneath all three.