菜单
Triage non-code defects (spec_defect, policy_incongruity, design_gap): assess risk, auto-fix bounded safe issues, and log the rest as tracked bugs.
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。
基于 SOC 职业分类
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.
Policy-as-code guardrail — loads an org policy bundle (file offline, world_tree online) and exposes a CHECK op used by the pre-tool-call and pre-commit hooks to deterministically block/allow. Team/Org tier only; free/retail installs run an empty/default bundle.
| name | g-skl-auto-triage |
| description | Triage non-code defects (spec_defect, policy_incongruity, design_gap): assess risk, auto-fix bounded safe issues, and log the rest as tracked bugs. |
Close the "detect and forget" loop. When a non-code defect is discovered (spec_defect, policy_incongruity, design_gap), this skill:
The goal is not to be brave. The goal is to leave fewer things silently broken.
subsystem_memberships: [BUG_AND_QUALITY, AGENT_ORCHESTRATION] --- risk — computes a numeric score; high risk = do not touch 2. FIXes bounded, safe issues automatically (score ≤ threshold) 3. LOGs every attempt to the audit trail regardless of outcome
The goal is not to be brave. The goal is to leave fewer things silently broken.
Compute risk score without writing anything.
ASSESS bug_id=BUG-098 kind=spec_defect files=["workspace_manifest.yaml"]
Returns:
risk_score: 1.5
eligible: true
reason: "spec_defect + single schema file + comment-only fix"
Attempt a bounded auto-fix. Only runs if risk_score ≤ auto_triage_risk_threshold (default: 2.0).
FIX bug_id=BUG-098 kind=spec_defect fix_type=schema_comment
Fix types available in Phase 1:
schema_comment — append clarifying comment to a YAML schema valuemanifest_annotation — add NOTE: inline to workspace_manifest.yamlcommand_annotation — add BUG[BUG-NNN] comment at bug site in a command filerule_annotation — add BUG[BUG-NNN] comment at bug site in a rule fileconstraint_expire — mark a constraint status: archived when expiry conditions are clearly metWrite an audit entry regardless of outcome. Called internally by FIX but can be called standalone.
LOG bug_id=BUG-098 outcome=auto_resolved risk_score=1.5 note="Added clarifying comment"
Appends to: .gald3r/logs/triage_auto_YYYYMMDD.log
risk_score = base_kind_score + file_sensitivity_bonus + scope_multiplier
| kind | base_score | Notes |
|---|---|---|
code | ∞ | NEVER auto-triage — code bugs go to normal fix path |
spec_defect | 1.0 | Safe starting point |
policy_incongruity | 2.0 | Borderline — only simplest fixes |
design_gap | 3.0 | Always needs_attention — human decision required |
| File type | Bonus |
|---|---|
| Schema YAML comment | +0.0 |
| Manifest annotation | +0.5 |
Rule file (.mdc) | +1.0 |
| Command file | +1.0 |
AGENT_CONFIG.md | +1.0 |
CONSTRAINTS.md body | +∞ (block) |
TASKS.md, tasks/ | +∞ (block) |
BUGS.md, bugs/ | +∞ (block) |
workspace/ topology | +∞ (block) |
| Any source code file | +∞ (block) |
| Files affected | Multiplier |
|---|---|
| 1 file | ×1.0 |
| 2–3 files | ×1.5 |
| >3 files | ×∞ (block) |
Threshold (Phase 1): risk_score ≤ 2.0 → attempt fix.
triage_status | Meaning | Bug badge |
|---|---|---|
auto_resolved | Fix applied cleanly | ✅ |
needs_attention | Human required (risk too high or fix failed) | 🚨 |
blocked_by_risk | Score exceeded threshold — logged, no write made | ⛔ |
deferred_verify | Fix applied, human should confirm before closing | 🔍 |
not_attempted | Triage not yet run for this bug | — |
kind: spec_defect # code | spec_defect | design_gap | policy_incongruity
triage_status: auto_resolved # not_attempted | auto_resolved | needs_attention | blocked_by_risk | deferred_verify
triage_risk_score: 1.5
triage_attempted: '2026-05-21T11:55:00Z'
triage_notes: 'Added clarifying comment to workspace_manifest.yaml line 1205'
TASKS.md, tasks/, BUGS.md, bugs/ — coordination state is sacredCONSTRAINTS.md body textworkspace/ topology filesdesign_gap)# Assess only
gald3r bug triage --dry-run --kind "spec_defect" --file "workspace_manifest.yaml" --fix-type "schema_comment"
# Full triage (assess + fix if safe + log)
gald3r bug triage --bug-id "BUG-098" --kind "spec_defect" --file "<workspace>/<gald3r_source>/.gald3r/linking/workspace_manifest.yaml" --fix-type "schema_comment" --project-root "<workspace>/<gald3r_source>"
2026-05-21T11:55:00Z | BUG-098 | spec_defect | risk=1.5 | auto_resolved | Added clarifying comment to workspace_manifest.yaml
2026-05-21T11:56:00Z | BUG-099 | policy_incongruity | risk=3.5 | blocked_by_risk | Score exceeds threshold 2.0
auto_triage_risk_threshold: 3.5)design_gap human-prompt mode (surfaces a proposed resolution, waits for approval)g-go-code as a pre-step impact classifier