Skip to content
Agent Memory
Command

/wiring-check

End-of-task wiring gate — verify every change is connected end-to-end across kipi plugins, hooks, MCP tools, agents, bus files, canonical, and rules. Nothing dangling.

BOOST
From plugin
kipi-system
12823 skills5 agents23 commands5 MCP
Install
> /plugin marketplace add assafkip/kipi-system

How it fires

How this command gets triggered: by you, by Claude, or both.

  • Fires itselfClaude auto-loads it when your prompt matches the work.
  • You can call itInvoke it directly when you want it.
  • Slash command/wiring-check

Context preview

What this command does when you run it.

End-of-task wiring gate — verify every change is connected end-to-end across kipi plugins, hooks, MCP tools, agents, bus files, canonical, and rules. Nothing dangling.

Command definition

wiring-check.md
description: End-of-task wiring gate — verify every change is connected end-to-end across kipi plugins, hooks, MCP tools, agents, bus files, canonical, and rules. Nothing dangling.
allowed-tools: Bash, Read, Grep, Glob

Run a Wiring Check on the work done in this session. Treat "this session" as the current `git diff` against the default branch unless the founder specifies otherwise (e.g., `since HEAD~3`, or a file list).

Goal: nothing disconnected, unused, unreachable, or half-integrated. If a file was created or edited, prove it is reachable from something that already runs.

Process (required, in order)

1. Enumerate changes

Run:

git diff --name-status origin/main...HEAD
git status --short

List every changed/created/deleted file. Group by kipi category (see matrix below).

2. Map to integration points

For each changed file, state the expected wiring based on its path. Use the matrix below. If a file doesn't fit any category, call it out — it may be in the wrong directory per `.claude/rules/folder-structure.md`.

3. Verify each wire

Use Grep/Glob/Read to prove the connection exists. No assumptions. If verification requires running a script, run it.

4. Fix gaps, then re-verify

If a wire is missing, fix it. Do not leave a "TODO: wire this up" — that is the exact failure mode this command exists to prevent.

5. Output WIRING REPORT

Structure:

  • **Changes shipped** (file list, grouped by category)
  • **Wiring evidence** (file + symbol/line for each connection)
  • **Verification run** (commands executed + results, pass/fail)
  • **Gaps found and fixed** (if any)
  • **Exceptions** (rare; rationale + tracking reference)
  • **Skeleton propagation status** — if this is the skeleton repo

(`kipi-system`), confirm `kipi update --dry` would pick up these changes; if this is an instance repo, note whether the change belongs upstream in the skeleton

Wiring Matrix (kipi-specific)

Check the wires that apply. Do not skip a category because "it looks obvious."

New or edited skill (`plugins/*/skills/<name>/SKILL.md`)

  • [ ] SKILL.md has frontmatter with `name`, `description`, `license`
  • [ ] Skill directory is under a plugin that ships (kipi-core / kipi-ops / kipi-design / kipi-dsse / prd-os)
  • [ ] Any references in the skill (scripts, refs/) exist at the stated paths
  • [ ] If auto-invoked, the trigger rule exists in `.claude/rules/*-auto-invoke.md` OR the skill description contains a clear trigger pattern
  • [ ] No duplicate skill name across plugin groups

New or edited slash command (`plugins/*/commands/<name>.md`)

  • [ ] Frontmatter has `description` and `allowed-tools`
  • [ ] Command name is listed in the project CLAUDE.md commands section (if project-wide)
  • [ ] Any scripts the command calls exist and are executable (`${CLAUDE_PLUGIN_ROOT}/scripts/...`)
  • [ ] Any templates/rubrics the command references exist

New or edited hook (`settings.json` hooks + hook scripts)

  • [ ] Hook script is referenced from `.claude/settings.json` (or plugin hooks config)
  • [ ] Script file exists, has `#!/usr/bin/env ...` shebang, and is `chmod +x`
  • [ ] Script uses `set -euo pipefail` (bash) or explicit error handling (Python)
  • [ ] Exit codes match the Claude Code hook contract (0 = ok, 2 = block with message)
  • [ ] Cross-check `.claude/rules/token-discipline.md` for any existing guard behavior that could conflict

New or edited MCP tool (kipi-core/kipi-mcp)

  • [ ] Tool is registered in the MCP server's tool list
  • [ ] Tool appears in the plugin description (so it's discoverable)
  • [ ] Any resource URIs (`kipi://...`) are served by the same server
  • [ ] Input schema is set; required params are marked

New or edited agent (`.claude/agents/*.md` OR `q-system/.q-system/agent-pipeline/agents/*.md`)

  • [ ] Frontmatter names a concrete model ID (per `memory/reference_agent_models.md` — do not use deprecated IDs)
  • [ ] Tool allowlist is explicit
  • [ ] For pipeline agents: input/output bus files match a schema in `agent-pipeline/schemas/`
  • [ ] Orchestrator invokes the agent in the expected phase

New or edited bus file / schema

  • [ ] `_bus-envelope.schema.json` structure is respected
  • [ ] Schema file exists at `agent-pipeline/schemas/<name>.schema.json`
  • [ ] At least one producer writes it AND at least one consumer reads it
  • [ ] `verify-bus.py` passes

New or edited canonical file (`q-system/canonical/*.md`)

  • [ ] Change is referenced in `ripple-graph.json`
  • [ ] `canonical-digest.py` was re-run and the bridge digest is updated
  • [ ] Council check fired (per `.claude/rules/auto-detection.md`) if >5 lines or new section

New or edited rule (`.claude/rules/<name>.md`)

  • [ ] Rule is auto-loaded (project-scoped rules in `.claude/rules/` load by convention) OR is imported via `@` in CLAUDE.md
  • [ ] Rule name doesn't shadow an existing one
  • [ ] If ENFORCED, the enforcement mechanism exists (hook, script, or explicit agent instruction)

New or edited Python harness (`q-system/.q-system/*.py` or `scripts/*.py`)

  • [ ] QROOT resolves to `q-system/` per folder-structure rule (`..` or `../..` depending on depth)
  • [ ] At least one caller references it (command, hook, or another script)
  • [ ] Runs clean on an empty/fresh instance (not just this founder's state)

New or edited PRD / issue spec (`q-system/output/prd-*.md`, DSSE issues)

  • [ ] PRD uses the template at `q-system/marketing/templates/prd.md`
  • [ ] Mandatory wiring checklist section is filled
  • [ ] If advancing state, `prd_runner.py` / DSSE scripts stamped the expected receipts

Skeleton → instance propagation (when working in `kipi-system`)

  • [ ] File is in a path `kipi-update.sh` copies (check the script's include list)
  • [ ] If it's instance-specific (e.g., `my-project/`, `memory/`, `output/`), it is NOT being shipped to the skeleton
  • [ ] Run `kipi update --dry` — the changed files should appear in the preview

Standard verification commands

Run whichever apply:

python3 q
Read more
Ships withkipi-system

Give your AI the right knowledge, at the moment it needs it. Kipi turns the scattered knowledge around your work into context that follows you across AI sessions: decisions, conversations, projects, people, commitments, lessons, documents, history.

Get the whole plugin

Other commands on kipi-system.