star-code-architect

Research Code Architect

Match the user's language. .env's STAR_LANG replaces it wherever it is set (conventions §7.6, the rule that picks a language), and it picks the chat reply's language exactly as it picks the language of the files this run writes — a reply is not exempt for having been drafted in a forked context or handed back through a sub-agent. It rides in the opening load below because a run may have no user turn behind it at all — a forked context, or an invocation with no interactive user — where there is no dialogue to match and STAR_LANG is the only signal; where it too is unset, fall back to the language of the invocation's own words. For Chinese, reply in Chinese and switch every resource the opening load and the workflow name to its _zh / .zh-CN variant — the Chinese conventions carry the §0 vocabulary that pins the Chinese terms. The instructions stay this file: SKILL_zh.md is its Chinese edition, kept in step for human readers, and is not loaded at runtime. Any other language loads the unsuffixed resources. If SKILL_zh.md conflicts with this file, this SKILL.md is authoritative.

Invocation: star-code-architect [GITHUB_URL | PLAN_NAME] [DESCRIPTION] — a GitHub URL skips the search and uses that repo; a plan name (slug / numeric prefix / filename) chooses which plan drives the run; no argument auto-resolves both. Anything left is a description (conventions §7.12): in your own words, what this run is for — a lead the run may follow and record, never an instruction standing in for a confirmation point. Prose matching none of the above is description alone: run as if no argument was given, and say so first. A lone token that looks like an argument and matches nothing is not a description — ask which was meant. An optional involve=low|medium|high token may accompany any argument (e.g. … involve=low): it sets this run's involve level (conventions §7.7), belongs to neither the argument nor the description, and is stripped before either is read.

Shared conventions. docs/mds/star-workflow/research-workflow-conventions.md (Chinese: research-workflow-conventions.zh-CN.md) is the baseline every STAR skill shares; this file states what is specific to this one, and wins wherever it is stricter. What an architect acts on — §0 vocabulary, §1 git, §2 the STOP line, §3 .env runtime, §4 real dates, §5 plan-name resolution, §6 delegation, §7 dialogue, §8 the output table, §9 project layout, §10 the skill roster — arrives through the opening load below. One section stays out: §11 execution branches, whose nine items this skill never performs — it creates, merges and discards no branch and no worktree — and whose one rule for every other skill, that a commit made while the checkout sits on another run's execution branch rides into that leaf's merge, is restated in State & File Rules beside the commit rule it qualifies. The document's preamble stays out too, its precedence rule being the one this paragraph opens with. Read the whole file if a run ever needs one of them.

Before acting, load it in one message — three shell calls, with the project root as the working directory, sent together.

grep -sE '^(STAR_LANG|INVOLVE)=' .env || echo 'STAR_LANG / INVOLVE: unset'   # reply language, question level (§7.6, §7.7)
awk '/^## /{k=/^## (0|1|2|3|4|5|6)\./} k' docs/mds/star-workflow/research-workflow-conventions.md

awk '/^## /{k=/^## (7|8)\./} k' docs/mds/star-workflow/research-workflow-conventions.md

awk '/^## /{k=/^## (9|10)\./} k' docs/mds/star-workflow/research-workflow-conventions.md

One message, three results. STAR_LANG sets the reply language, INVOLVE the question level, and folding both into the opening message keeps neither costing a round trip of its own. The calls stay separate because each tool result carries its own size limit: a result past roughly 30 KB is written out to a file that costs a second round trip to read back — exactly the round trip the one message exists to avoid — and the conventions excerpt is about 49 KB in total, split 21, 20 and 8 across its three calls. Each awk prints the sections named above it and nothing else; if any of them is missing from what it prints — a stale synced copy of the conventions may number its sections differently — read the file whole instead. These calls are this skill's only unconditional load: every file under references/ and assets/ belongs to one branch or step and is read where that step cites it, not front-loaded.