diagnose-substrate

스타9

포크1

업데이트2026년 5월 21일 01:06

Audit an existing Claude Code workspace for canon-substrate gaps — the failure mode where users build sophisticated agent stacks without a CORPUS_MAP / SCHEMA / decisions/ layer underneath, leaving every agent's "consult on demand" reference list resolving to nothing. Reports ship-blocking issues, fragile-but-working issues, and noise. Use when invited into someone's `.claude/agents/` directory to do an audit-and-recommend pass.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

출처

liminalshruti

liminalshruti/liminal-agents

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Liminal Agents — Substrate Diagnostic

You are auditing a Claude Code workspace for canon-substrate gaps. The workspace may have an existing agent layer (subagents in .claude/agents/) that runs on prompt-only context with no canonical substrate underneath. This audit identifies what's broken, fragile, or noisy — and proposes the foundation pack shape that would fix it.

The audit is read-only. It does NOT modify the workspace. It produces a structured report.

When to invoke

User has a working .claude/agents/ directory and asks "is my setup right?"
User shares a snapshot zip of their Claude config and wants a review
User reports "Claude doesn't remember things between sessions" — high-probability symptom of memory-path mismatch or empty memory dir
Before offering canon-substrate templates from liminal-agents/templates/canon-substrate/, run this diagnostic first so the templates are tuned to actual gaps

Flow

1. Validate input

User provides a workspace root path. Verify:

Path exists and is a directory
.claude/agents/ exists under it (or .claude/ somewhere reachable)
If neither, abort with "no agent layer found at /.claude/agents/"

2. Run the diagnostic

Invoke:

node ~/liminal/liminal-agents/skills/diagnose-substrate/diagnose.js <workspace-path>

The script produces a JSON report with sections:

broken[] — ship-blocking issues (memory-path mismatch, missing CLAUDE.md, no CORPUS_MAP)
fragile[] — works but risky (quarantine model relies only on system-prompt-only declarations, settings flags that disable safety prompts, agent overlap)
noise[] — cosmetic issues (inert JSON-comment-string entries in allow arrays, etc.)
agent_summary — count of agents, average lines, models used, tools granted
recommended_foundation_pack — which template files would address each broken/fragile issue

3. Present the report

Render the report in this order:

Headline — one line summarizing severity (e.g. "3 ship-blocking, 4 fragile, 1 noise")
Broken issues — for each, name the issue + the canonical fix
Fragile issues — for each, name the issue + the risk vector
Noise — flagged but not blocking
Foundation-pack recommendation — link to liminal-agents/templates/canon-substrate/ with which template files address which gaps

4. Offer next step

Ask the user: would they like the foundation pack deployed to their workspace? If yes, do NOT auto-deploy — produce a paste-into-Claude landing prompt (similar to the one Shruti shipped to Sam on 2026-05-19) that lands the templates without disturbing the existing agent layer.

Output principles

Read-only. Never modify the workspace.
Honest severity. Don't soften broken-→-fragile. Memory-path mismatch is broken, even if "the agents still respond." They respond on empty context.
No upsell. The diagnostic is the deliverable. Foundation-pack deployment is a separate action, surfaced only if the user asks.
Specific, not generic. "Your memory dir resolves to ; the agents reference ; the difference is the project-name mismatch from a prior CWD" — not "your memory might not work."

Related canon

Architectural rationale: liminal-agents/SPEC.md §14 "Canon Substrate as Agent-Layer Prerequisite"
Foundation pack templates: liminal-agents/templates/canon-substrate/
Reference instance (Shruti's own canon): ~/liminal/founder-brain/CLAUDE.md + ~/liminal/founder-brain/meta/CORPUS_MAP.md
Original live audit that produced this skill: ~/liminal/founder-brain/retros/2026-05-19-helping-sam-shape-of-help.md

이 저장소의 다른 Skills

같은 저장소

govern

liminalshruti/liminal-agents

Govern AI spend without surveilling your team — the AI-spend governance loop. Set OKRs, harden the right context (tray→slate), bounded agents surface a finding ready for ratification, ratify a policy to the provenance trail, get a spend-ROI report that cites its sources. The leader governs usage; the team's private context is never read.

2026-06-229

onboard-swarm

liminalshruti/liminal-agents

The parallel per-agent swarm that beats cold-start. After install, each bounded agent reads its canonical source (git/claude-code → Analyst, granola → SDR, cross-stream → Auditor) in parallel and posts a candidate — the one fact, commitment, or risk worth acting on — before any deliberation. Read-only, LLM-optional. Run this to show the vault won't start empty.

2026-06-139

check

liminalshruti/liminal-agents

Twelve-agent deliberation on the user's psychological state across four registers (Structural / Somatic / Temporal / Symbolic). Asks three forced-choice questions, runs all 12 agents via Opus 4.7 in parallel, surfaces their disagreement grouped by register, and writes the user's correction — tagged from the canonical taxonomy — to the local vault.

2026-06-139

try-liminal

liminalshruti/liminal-agents

The 60-second taste of Liminal. Runs one bounded deliberation on a sample brief — three agents read the same brief and disagree, at least one refuses out of its lane, you correct one read, and the correction becomes the record. Ends by pointing at the desktop app so the vault outlives the session. Run this first.

2026-06-139

scaffold-substrate

liminalshruti/liminal-agents

Materialize the canon-substrate foundation pack into a target workspace. Reads a diagnose-substrate JSON report + a small config file, copies the recommended template files into the target, substitutes high-frequency placeholders (PROJECT NAME, OWNER NAME, sensitivity tiers), and emits a landing prompt the user pastes into their Claude Code session to finish the customization. Pairs with diagnose-substrate — diagnose first, scaffold second. Step (b) of SPEC §14's productized Diagnostic + Foundation Pack pattern.

2026-05-219

forensic

liminalshruti/liminal-agents

Run the Forensic agent against an inbound artifact (.eml or .pdf) — extracts provenance, authentication state, CRM internal IDs, recipient routing, UUIDs in body text — and returns structured receipts. The Forensic agent refuses interpretation, verdict, synthesis, and action-proposal — those route to Auditor / Skeptic / Strategist / Operator if the user invokes them separately.

2026-05-199

name	diagnose-substrate
description	Audit an existing Claude Code workspace for canon-substrate gaps — the failure mode where users build sophisticated agent stacks without a CORPUS_MAP / SCHEMA / decisions/ layer underneath, leaving every agent's "consult on demand" reference list resolving to nothing. Reports ship-blocking issues, fragile-but-working issues, and noise. Use when invited into someone's `.claude/agents/` directory to do an audit-and-recommend pass.
disable-model-invocation	true
allowed-tools	Bash(node *)
argument-hint	<path to workspace root containing .claude/agents/>

Liminal Agents — Substrate Diagnostic

The audit is read-only. It does NOT modify the workspace. It produces a structured report.

When to invoke

User has a working .claude/agents/ directory and asks "is my setup right?"
User shares a snapshot zip of their Claude config and wants a review
User reports "Claude doesn't remember things between sessions" — high-probability symptom of memory-path mismatch or empty memory dir
Before offering canon-substrate templates from liminal-agents/templates/canon-substrate/, run this diagnostic first so the templates are tuned to actual gaps

Flow

1. Validate input

User provides a workspace root path. Verify:

Path exists and is a directory
.claude/agents/ exists under it (or .claude/ somewhere reachable)
If neither, abort with "no agent layer found at /.claude/agents/"

2. Run the diagnostic

Invoke:

node ~/liminal/liminal-agents/skills/diagnose-substrate/diagnose.js <workspace-path>

The script produces a JSON report with sections:

broken[] — ship-blocking issues (memory-path mismatch, missing CLAUDE.md, no CORPUS_MAP)
fragile[] — works but risky (quarantine model relies only on system-prompt-only declarations, settings flags that disable safety prompts, agent overlap)
noise[] — cosmetic issues (inert JSON-comment-string entries in allow arrays, etc.)
agent_summary — count of agents, average lines, models used, tools granted
recommended_foundation_pack — which template files would address each broken/fragile issue

3. Present the report

Render the report in this order:

Headline — one line summarizing severity (e.g. "3 ship-blocking, 4 fragile, 1 noise")
Broken issues — for each, name the issue + the canonical fix
Fragile issues — for each, name the issue + the risk vector
Noise — flagged but not blocking
Foundation-pack recommendation — link to liminal-agents/templates/canon-substrate/ with which template files address which gaps

4. Offer next step

Output principles

Read-only. Never modify the workspace.
Honest severity. Don't soften broken-→-fragile. Memory-path mismatch is broken, even if "the agents still respond." They respond on empty context.
No upsell. The diagnostic is the deliverable. Foundation-pack deployment is a separate action, surfaced only if the user asks.
Specific, not generic. "Your memory dir resolves to ; the agents reference ; the difference is the project-name mismatch from a prior CWD" — not "your memory might not work."

Related canon

Architectural rationale: liminal-agents/SPEC.md §14 "Canon Substrate as Agent-Layer Prerequisite"
Foundation pack templates: liminal-agents/templates/canon-substrate/
Reference instance (Shruti's own canon): ~/liminal/founder-brain/CLAUDE.md + ~/liminal/founder-brain/meta/CORPUS_MAP.md
Original live audit that produced this skill: ~/liminal/founder-brain/retros/2026-05-19-helping-sam-shape-of-help.md