메뉴
Reads the tracker (GitHub Issues), picks the pinned/top open issue (or the issue the user names), and restates it as a concrete in-session work plan naming the files and crates involved. Use when the user asks "what's next" or "what should I work on".
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Closes a tracker issue: verifies the work is genuinely done (exit criteria, gates, docs, changelog), writes the close narrative into the PR description, ensures the PR declares `Closes #N`, posts the handoff comment, and deletes the implemented plan file. Use when the user says a work item is complete or asks to close it out.
Runs the CNF 2.0 conformance pipeline (scripts/conformance.sh) — the reference runner over the committed machine-readable catalogue and the acceptance instrument — against the Docker-composed server (or any BYO SUT). Use when the user asks to check conformance, run the conformance/CNF suite, verify spec compliance, or at phase close for the zero-drift gate.
Audits an implemented subsystem, endpoint group, or diff against the vendored openEHR spec text and the CNF Platform Conformance Test Schedule (docs/specs/openehr/), producing a findings list of divergences with spec citations. Use when the user asks "are we following the spec", "audit X against the spec/CNF", before closing a spec-facing phase, or after a conformance failure.
Prints the tracker's open issues (pinned first), any linked plan file's unchecked tasks, and a short git status. Use when the user asks "where are we", "what's in flight", or at the start of a work session to orient.
Creates a new workspace crate directory (crates/, app/, or tools/) with a Cargo.toml wired to the workspace (inherited package fields, [lints] workspace = true, path deps per the docs/architecture.md dependency arrows), a doc-comment-only lib.rs/main.rs, and the crate's nested CLAUDE.md. Use when a plan calls for standing up a crate that does not exist yet.
Regenerates the openEHR spec + ITS layer from the vendored specs (emit / emit-xml / emit-rest) and checks for drift. Use after changing the openehr-codegen emitter or bumping a vendored spec, or when asked to regenerate the generated crates or run the codegen drift check.
| name | next-task |
| description | Reads the tracker (GitHub Issues), picks the pinned/top open issue (or the issue the user names), and restates it as a concrete in-session work plan naming the files and crates involved. Use when the user asks "what's next" or "what should I work on". |
| allowed-tools | ["Read","Grep","Glob","Bash"] |
| argument-hint | [issue number] (optional) |
Turns an open tracker issue into an actionable plan — the planning step of
the issue workflow (CLAUDE.md). Does not do the work itself; that is a
separate step the caller takes after seeing the plan.
gh issue list --state open (the SessionStart dump
annotates each issue with {k/n} sub-issue progress, child-of #parent,
and open BLOCKED-by/blocks edges) — take the pinned issue (current
focus) or the issue the user named. Respect relationships
(.claude/rules/issue-relationships.md): do NOT pick an issue shown
BLOCKED-by an open issue (surface its blocker as the real next task
instead); for a parent issue, point at its next open child rather than the
parent itself. Then gh issue view <n> --comments for the full contract
(## Contract + ## Exit criteria) and the running discussion, and
scripts/gh-rel.sh tree <n> for its parent/children/blockers. If the issue
links a plan file (docs/plans/*.md), read that too; its unchecked
(- [ ]) tasks are the queue.crates/ and app/) rather than guessing paths; if the task names a
spec component, resolve it against docs/architecture.md (the crate map)
and docs/specs/openehr/README.md.openehr-base/openehr-rm/openehr-am/
openehr-its) → the code generator — change openehr-codegen's
emitter and regenerate (/regen-codegen).
Application (ehrbase/ehrbase-rest) → idiomatic
modern Rust of our own design on the generated crates, the openEHR
specs as the authority (EHRbase = prior art only). Build
compiling + tested. Note whether the task suits fanning out to an
implementer/ui-implementer subagent per the CLAUDE.md
Model-orchestration section (max 2 concurrent), or belongs in-session
(architecture, the AQL IR/codec core, spec-conformance judgement).docs/specs/openehr/... files (and CNF test-schedule chapters) the
implementation must be read against, per spec-adherence.md /
/spec-lookup. Doing the work starts by reading those.## Exit criteria checklist, plus what proves it: the CNF pipeline's
zero-drift gate (docs/conformance/ehrbase-rs/results.json +
verdicts.json), the openehr-its fidelity gates, or corpus tests.