archi-silent-audit

name	archi-silent-audit
description	Lightweight code and document review. Must run in isolated context/subagent. Protocol-invoked only; do not auto-trigger from casual user requests.
disable-model-invocation	true

Invocation

• Auto-invoke: No, not triggered by model based on description.
• Trigger location: Only explicitly called via [[SUBAGENT]] / [[NO-SUBAGENT]] in /archi.* protocols.
• Execution context: When subagent supported must execute in independent subagent/independent context; only downgrade to inline Skill when no subagent.
• Boundary: Only return protocol-required structured artifacts, subsequent write, confirm and signoff handled by calling protocol.

Core Principles

• No Evidence No Report: Each finding must have evidence (original snippet + location); guess without locatable evidence never reported
• Pre-bound Level: severity defined by dimension library table below, skill only judges "Triggered / Not triggered" not level
• All Open Once: Dimensions filtered by mode run all at once, not staged, not early-stop
• Level ≠ Action: Findings and levels returned as-is, how to respond decided by calling protocol

Dimension Library

Symbol legend:

• Mode column: I = init, P = plan-docs, C = code-impl (Multiple use / separator)
• Gate column: [[WHEN: X]] = Deploy-time features filter; runtime: X = Judgment at call-time via task_meta; — = No gate

#	ID	Name	Mode	Level	Gate	Trigger Condition
1	VISION_ROADMAP_ALIGN	Vision-Roadmap Alignment	I	CRITICAL	—	roadmap task goal direction inconsistent with vision.md north star direction (task doesn't serve north star)
2	TECH_STACK_DECLARED	Tech Stack Declaration Consistent	I / P / C	CRITICAL	—	init: tech_stack.md doesn't reflect package.json actual dependencies; plan-docs: spec/plan uses library/framework/API not declared in tech_stack; code-impl: code imports library not in tech_stack
3	GLOBAL_FILES_COMPLETE	Global Files Completeness	I	CRITICAL	—	vision / roadmap / map / dictionary / tech_stack / custom_rules any missing or only skeleton without substance
4	INFORMATION_ROUTING	Information No Leakage	I	WARNING	—	Information in Brief / code (feature points / terms / constraints) not routed to corresponding global files
5	DESIGN_TOKENS_BASIC	Design Tokens Basics Ready	I	CRITICAL	[[WHEN: ui]]	design_tokens.json missing aestheticDirection.preset / primitivePalette.brand / core font size fields
6	DESIGN_FIDELITY	Design Fidelity	P	CRITICAL	—	spec § 2 doesn't cover feature design points confirmed in discussion (has omissions)
7	DIMENSION_MATCH	Dimension Match	P	WARNING	—	spec § 2 dimension format doesn't match Task Type (didn't output per Task Type specified dimensions)
8	WBS_COVERAGE	WBS Coverage	P	CRITICAL	—	plan.json has spec AC entries not covered by any task
9	NOTES_QUALITY	Notes Quality	P	WARNING	—	plan.json task.notes lacks "Output:" / "Constraint:" / "Verification:" any keyword; or "Verification:" value contains "appropriately", "reasonably", "as needed", "per situation" etc vague quantifiers
10	INTERFACE_EXPORTS	Interface Exports	P	CRITICAL	—	INF task § 4 not filled; or downstream deps exist but interface signature not declared
11	CONSTRAINTS_REDLINES	Constraints Red Lines	P	WARNING	—	spec § 5 lacks red lines related to vision.md north star or tech_stack known prohibited items
12	DATA_INTEGRITY_SPEC	Data Integrity (spec)	P	CRITICAL	[[WHEN: data]]	spec entity/field inconsistent with data_snapshot.json existing core entity/field
13	DESIGN_TRACE	Design Trace	P	WARNING	runtime: is_complex	design.md § 6 trace table has AC not mapped to design node
14	PARAMETER_SPECIFICITY	Parameter Specificity	P	INFO	runtime: is_complex	design.md § 3 parameter table has "appropriately", "reasonably", "as needed", "per situation", "some" etc vague quantifiers
15	DESIGN_SELF_CHECK	Design Self-Check	P	CRITICAL	runtime: is_complex	design.md § 2 mechanism self-check list any item marked ✗ or empty unchecked
15.5	TEST_PLAN_QUALITY	Test Plan Quality	P	WARNING	—	plan.json tests / task notes only write "add tests", "run tests", "cover logic" etc placeholder sentences; or not mapped to spec AC / bug reproduction steps
16	TECH_STACK_STYLE_MATCH	Tech Stack Style Consistent	C	WARNING	—	Code pattern / API style deviates from tech_stack.md declaration (e.g. declared ESM but uses CommonJS, declared hooks-first but uses class component)
17	SOTA	SOTA	C	WARNING	—	Appears in tech_stack explicitly listed anti-pattern; or uses deprecated API replaced by same stack
18	SECURITY	Security	C	CRITICAL	—	Code has hardcoded secrets / plaintext passwords; or external input unvalidated directly into DB / FS / shell / eval
19	PERFORMANCE	Performance	C	WARNING	—	Obvious full imports, repeated computation, unreleased listeners, O(n²) optimizable to O(n) scenarios
20	UI_TOKEN_COMPLIANCE	UI Token Compliance	C	CRITICAL	[[WHEN: ui]] + runtime: involves_ui	Style values have hardcoded color / font size / spacing (not var(--*) or token reference)
21	ACCESSIBILITY	Accessibility	C	WARNING	[[WHEN: ui]] + runtime: involves_ui	Interactive elements lack aria-label / semantic tags / keyboard reachability (button / input / link missing required attributes)
22	DATA_INTEGRITY_CODE	Data Integrity (code)	C	CRITICAL	[[WHEN: data]] + runtime: involves_data	Code implemented field name / type inconsistent with data_snapshot.json
23	DESIGN_COMPLIANCE_STATE	Design Compliance State	C	CRITICAL	runtime: is_complex	Code state transition / flow / protocol inconsistent with design.md § 2
24	INVARIANT_ENFORCEMENT	Invariant Enforcement	C	CRITICAL	runtime: is_complex	design.md § 4 invariants not enforced in code via assert / runtime check / type guard
25	PARAMETER_ALIGNMENT	Parameter Alignment	C	CRITICAL	runtime: is_complex	Code values / thresholds inconsistent with design.md § 3 parameter table
26	TEST_BEHAVIOR_QUALITY	Test Behavior Quality	C	WARNING	—	New/modified tests only assert mock call counts, file exists, object non-empty, JSON parseable; lacking behavior results, boundary/error paths or Red→Green reproduction evidence

Execution Protocol

1. Filter dimensions with corresponding letter in "Mode" column by mode
2. Then filter by "Gate" column: Dimensions for features not present in current context视为inapplicable; runtime: judge via task_meta on the spot
3. For each retained dimension, scan corresponding content in context_files per "Trigger Condition"
   • In mode: init, only audit files generated/overwritten by this init run, or files explicitly returned in step_3 writtenFiles.
   • If a file still contains architextTemplate: true, Status: Template, template-uninitialized, or lastUpdated: "TEMPLATE", it is a scaffold seed, not an audit target; do not report roadmap/source mismatch, incomprehensible vision, or global document drift from it.
4. Hit → Produce finding (with dimension / location / evidence / description); No hit → No output
5. After all dimensions run, group output by severity

Output Format

===

Silent Audit Results (mode: )

CRITICAL:

• dimension: location: <file:line or § number> evidence: "" description:

WARNING:

• dimension: location: ... evidence: "..." description: ...

INFO:

• dimension: location: ... evidence: "..." description: ...

Summary: CRITICAL / WARNING / INFO

No findings only output: ### Silent Audit Results (mode: <mode>) — ALL PASS

Output Verification

• Each finding's evidence non-empty, can be located in context_files
• Each finding's severity matches pre-bound level in dimension library
• Dimensions not filtered by current mode don't appear in findings
• Dimensions for runtime gates not opened per task_meta don't appear (e.g. is_complex = false doesn't report Complex-type)
• Summary count matches actual finding list count