Menü
Use after repo-init-gate fails, or when explicit reinitialization/repair is requested, to orchestrate the official init stage and the manual fallback stage.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
| name | repo-init-scan |
| description | Use after repo-init-gate fails, or when explicit reinitialization/repair is requested, to orchestrate the official init stage and the manual fallback stage. |
Use this thin orchestrator when the repository needs full init or scaffold repair.
When this skill has just been loaded after repo-init-gate failed, the very next assistant action must be staged init work or a BLOCKED init report:
repo-init-gate/scripts/run-preflight.sh <target-root> claude; it preserves the sentinel fast path and delegates here only on a failed gate. If this scan skill has already been loaded after a failed gate, resolve the helper with CLAUDE_SKILL_DIR when present, otherwise use the active plugin directory when available and run <plugin-dir>/skills/repo-init-scan/scripts/bootstrap-after-gate-failure.sh <target-root> claude. This command owns the official attempt, deterministic manual fallback helper, disk verification, sentinel rewrite, and ## Init Summary output. If the helper path is not discoverable, continue to the strict inline fallback below instead of blocking solely on helper discovery.repo-init-official first, or execute its documented official-init fallback inline when slash skill execution is unavailable.repo-init-manual-fallback, including its bundled deterministic scaffold helper when shell access is available, or execute its documented scaffold verification and repair inline.repo-init-manual-fallback on disk, rewrite the exact best-copilot.md sentinel when needed, and emit ## Init Summary.Do not hand-create a shortened scaffold, search/read business source, inspect modules such as core, call code intelligence, plan, dispatch, or summarize implementation after a Skill(...repo-init-scan...) Successfully loaded line and before ## Init Summary. That transcript is invalid; recover by ignoring the premature context and executing the single-command bootstrap path, strict 17-path inline fallback, or staged init flow now. If file read/write or disk verification cannot actually run, return BLOCKED tool_execution_unavailable; never produce a prose-only success summary.
repo-init-gate first, then load this skill only when the gate reports needs_init, version_mismatch, or invalid_sentinel.best-copilot.md and return early when the YAML frontmatter version is already current.repo-init-official first, then repo-init-manual-fallback.best-copilot.md has been rewritten for version 0.7.1.Skill(best-copilot:repo-init-scan) Successfully loaded is only a preload trace. A valid INIT_SCAN requires running the preflight or scan bootstrap helper, or running the stages below inline, creating or repairing target-local files when needed, checking them on disk, and returning the init summary.repo-init-gate/scripts/run-preflight.sh or the bundled repo-init-scan/scripts/bootstrap-after-gate-failure.sh command. If shell execution is blocked but file tools work, a strict inline fallback is valid only when all Claude-compatible paths are created/repaired and verified. A hand-written two-file or four-file scaffold is invalid.required_artifacts_verified is not yes.BLOCKED init report. A code intelligence call, business-source read/search, architecture summary, or specialist dispatch before the init summary is invalid.## Repo Init Gate first, then real stage work, then the structured ## Init Summary fields below. A narrative claim that files were created without verified paths is invalid.required_artifacts_verified: yes is forbidden unless verified_paths covers every required artifact from repo-init-manual-fallback and missing_paths is none. A partial verified list means required_artifacts_verified: no, next_task_ready: no, and BLOCKED first_use_gate_incomplete.For required_artifacts_verified: yes, verified_paths must include these exact target-root paths. Similar paths do not count.
Base first-use init has 16 required paths including the sentinel. Claude-compatible first-use init has 17 required paths. Codex-compatible first-use init has 25 required paths because Codex custom-agent adapters are target-local .codex/agents/*.toml files. Fewer verified paths means required_artifacts_verified: no, even if best-copilot.md and .github/instructions/project.instructions.md exist.
.github/instructions/project.instructions.md.github/instructions/must.instructions.md.github/instructions/skills-index.instructions.mdCLAUDE.md when the active target runtime is Claude Code or Claude compatibility is requestedAGENTS.md when the active target runtime is Codex or Codex compatibility is requested.codex/agents/senior-project-expert.toml when Codex compatibility is requested.codex/agents/technical-architect.toml when Codex compatibility is requested.codex/agents/developer.toml when Codex compatibility is requested.codex/agents/frontend-designer.toml when Codex compatibility is requested.codex/agents/quality-assurance-expert.toml when Codex compatibility is requested.codex/agents/security-reviewer.toml when Codex compatibility is requested.codex/agents/specification-writer.toml when Codex compatibility is requested.codex/agents/root-cause-fixer.toml when Codex compatibility is requestedmemories/README.mdmemories/repo/INDEX.mdmemories/repo/current-workstreams.mdmemories/repo/project-state.mdmemories/repo/workflow-rules.mdmemories/repo/decisions.mdmemories/repo/logs/README.mdmemories/repo/archive/deprecated-decisions.mdspec/INDEX.mdspec/templates/requirements-template.mdspec/templates/design-template.mdspec/templates/tasks-template.mdbest-copilot.md after the sentinel rewriteDo not accept substitutes such as memory/current-state.md, MEMORY.md, spec/requirements.md, or spec/templates/spec-template.md. If any exact required path is absent or content-invalid, set missing_paths to those paths, required_artifacts_verified: no, next_task_ready: no, and return BLOCKED first_use_gate_incomplete.
The verified_paths output field must enumerate the exact relative path names. A phrase such as all required paths, created successfully, or an absolute temp directory path is not valid evidence.
When performing the strict inline fallback (no shell helper available), the exact content for target-root best-copilot.md is:
---
version: "0.7.1"
---
Do NOT add headings, project names, descriptions, dates, or any other prose when writing or repairing the sentinel. The canonical write format is only the YAML frontmatter block above. Repeat-request gate reads accept a current frontmatter version as ready so harmless extra content does not trigger a second full scan.
repo-init-official owns the target-root official initializer attempt: target-local init skill first when discoverable and mechanically invokable, then Claude native /init or Copilot copilot init command fallback, plus normalization of official output into .github/instructions/project.instructions.md when possible. In Claude Code, /init means Claude Code's native command shown as "Initialize a new CLAUDE.md file with codebase documentation", not a best-copilot skill. In Copilot CLI, copilot init is the native official initializer when the command exists; a target skills/init/SKILL.md is invokable as a Copilot local plugin only when target root also has plugin.json. A helper-created CLAUDE.md, .github/copilot-instructions.md, .github/instructions/project.instructions.md, or AGENTS.md is also a valid official artifact for repo-init-manual-fallback to preserve and normalize.repo-init-manual-fallback owns bounded manual scanning, deterministic scaffold bootstrap, artifact verification, and writing best-copilot.md.repo-init-gate fails, the manual fallback stage still owns the final scaffold verification barrier and sentinel rewrite./init command from best-copilot plugin skills. The official stage must run repo-init-official/scripts/run-claude-native-init.sh; that helper tries a target-local init skill first when present, then Claude's native initializer, before falling back to bounded manual scanning.repo-init-gate, do one shallow read of the target root best-copilot.md first:
INIT_SCAN=SKIP_SENTINEL_READY, stop, and treat repo init as already verified by the sentinel contract.repo-init-official first.repo-init-manual-fallback after the official stage:
best-copilot.md.best-copilot.md.CLAUDE_SKILL_DIR or the active plugin directory when available. Otherwise invoke the best-copilot stage skills when available, using the exact picker value such as /repo-init-official (best-copilot) or /best-copilot:repo-init-official, then /repo-init-manual-fallback (best-copilot) or /best-copilot:repo-init-manual-fallback. These plugin skills are different from the bare built-in /init command. If Skill(...) Successfully loaded appears but no actual init operations happen, that means the skill text loaded but was not executed — execute the documented steps inline now. If shell access exists and the manual fallback helper is discoverable, run its bundled scripts/bootstrap-target-scaffold.sh helper before any hand-written scaffold attempt. If the runtime only loads the skill text or cannot invoke plugin slash commands or discover the helper, execute the documented fallback inline: apply the official stage rules, then read the bootstrap skill templates and create/repair the target files directly..github/instructions/project.instructions.md, all required scaffolds, and a rewritten best-copilot.md sentinel for version 0.7.1.BLOCKED instead of continuing to the user's substantive task.Return a short initialization report:
## Init Summary
- official_stage: success|official_init_unavailable|official_init_no_write|official_init_incomplete
- official_attempted: target_local_init_skill|copilot_init|claude_native_slash_init|bare_init|none
- manual_fallback_stage: skipped|success|blocked
- required_artifacts_verified: yes|no
- sentinel_written: yes|no
- next_task_ready: yes|no
- verified_paths: <every required artifact from repo-init-manual-fallback checked on disk, plus best-copilot.md>
- missing_paths: none|<paths not present or not content-valid>
official_attempted may be a comma-separated chain when a target-local init skill exists but does not write a recognized artifact and the helper continues to the runtime command fallback.
sentinel_written: yes means the current best-copilot.md sentinel is present and current, whether it was newly written or already present. required_artifacts_verified: yes may be used only when missing_paths is none and verified_paths covers the full Required Artifact Set above, including Claude CLAUDE.md when Claude compatibility is active and Codex AGENTS.md plus .codex/agents/*.toml when Codex compatibility is active. next_task_ready may be yes only when required_artifacts_verified is yes and sentinel_written is yes. If any path is missing or unverified, the PM must stop at the first-use gate and return BLOCKED first_use_gate_incomplete, not a normal plan or implementation summary.
manual_fallback_stage: skipped is allowed only when this skill was called directly and the current sentinel already matched before any gate failure. After repo-init-gate reports a missing, invalid, or stale sentinel, the manual fallback stage must run because it owns scaffold verification and sentinel rewrite.
Use when an agent needs the shared best-copilot contract for source priority, runtime adapters, init gates, work modes, dispatch packets, review, verification, memory, spec, state persistence, or closeout.
Use inside repo-init-scan after the official stage whenever scaffold verification, bounded repair, or `best-copilot.md` rewrite is still needed.
Use when coordinating large, ambiguous, cross-module, multi-agent, planning, dispatch, fan-in, closeout, or workflow-evolution work as the Senior Project Expert.
Use when maintaining requirements, design, tasks, ADRs, execution-plan state, closeout records, or memory/spec recovery as the Specification Writer.
Use when an approved tasks.md from a Spec Bundle, or a compact approved small-work plan, should be executed by fresh-context specialist subagents, with each task receiving implementation, spec-compliance review, code-quality review, and verification before closure. DO NOT USE FOR: missing plans, brainstorming, or simple single-file edits.
Create the target repository's local spec skeleton and reusable requirements/design/tasks templates. Use from repo-init-scan during first substantial plugin use, or from spec-writing flows when `spec/INDEX.md` or `spec/templates` is missing. Do not overwrite active project specs.