メニュー
Own and manage all task data — TASKS.md index, tasks/ individual files, status transitions, sync validation, complexity scoring, and sprint planning. Single source of truth for everything task-related.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Triage non-code defects (spec_defect, policy_incongruity, design_gap): assess risk, auto-fix bounded safe issues, and log the rest as tracked bugs.
Compress the NON-gald3r sections of AGENTS.md/CLAUDE.md (and *memory*.md) to cut token overhead, while strictly preserving the install-managed gald3r SECTION ranges, code blocks, and URLs. Dry-run by default; apply only after confirmation.
Own and manage all feature data — FEATURES.md index, features/ individual files, staging lifecycle (staging→specced→committed→shipped), harvest source collection, and feature promotion. Single source of truth for everything feature-related.
Create well-structured git commits following gald3r conventions, with proper type prefixes, task references, and clean trailers (no AI co-author footers per C-021).
Tiered .gald3r/ health and intervention system. L1=triage (structural), L2=diagnosis (plan coherence), L3=surgery (cross-subsystem interface audit), L4=ecosystem (linked project negotiation). Replaces g-skl-medic.
gald3r_muninn MCP — query the local codebase knowledge graph for impact analysis, caller chains, dependencies, and symbol search. Clean-room rewrite (T1147 epic, T1153-T1158) of the GitNexus integration; auto-loaded by the example_app MCP server from docker/gald3r/tools/plugins/muninn/. Wire into g-go-code Step b0 Impact Scan before any implementation.
| name | g-skl-tasks |
| maturity | production |
| description | Own and manage all task data — TASKS.md index, tasks/ individual files, status transitions, sync validation, complexity scoring, and sprint planning. Single source of truth for everything task-related. |
| token_budget | low |
| subsystem_memberships | ["TASK_MANAGEMENT"] |
Handled by the bundled gald3r engine (
.gald3r_sys/engine, pure Mode-A, no LLM). Full original procedure retained inSKILL.full.mdso an install without the engine still works.
What it does: task lifecycle over TASKS.md + tasks//.
gald3r task … (or the installed gald3r)gald3r_task_* · facade Gald3r(...).tasksThe engine owns ID allocation, file placement, status→folder moves, index regeneration, and
validation. .gald3r/ markdown stays the data source of truth.
During a multi-agent / autopilot run (marker .gald3r/logs/ggo_run_state.json active: true,
or env GALD3R_AGENT_RUN=1), the engine create() auto-routes a new task to tasks/inbox/
as an id-less draft (uuid-suffixed filename) instead of assigning an id directly. The
hot-inbox intake (run at each iteration boundary) is the single ID-assigning authority and
assigns ids atomically — so concurrent agents can never collide on the next id. Idle (no run) →
direct create, unchanged. Hand-writing agents (manual fallback) MUST drop new-task drafts in
tasks/inbox/ during a run — never write tasks/open/ + regenerate the index directly.
Follow SKILL.full.md (full procedure) + the schema in .gald3r_sys/schemas/ (task-file).
Everything needed ships in the install — nothing external.