@webpresso/claude-plugin 3.1.17 → 3.1.19

package/.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
   },
   "metadata": {
     "description": "Webpresso agent-kit Claude Code plugin: blueprints, skills, hooks, MCP server",
-    "version": "3.1.17"
+    "version": "3.1.19"
   },
   "plugins": [
     {
@@ -34,5 +34,5 @@
       }
     }
   ],
-  "version": "3.1.17"
+  "version": "3.1.19"
 }

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-kit",
-  "version": "3.1.17",
+  "version": "3.1.19",
   "description": "Webpresso agent-kit: blueprints, skills, lore commit protocol, tech-debt lifecycle",
   "author": {
     "name": "Webpresso",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@webpresso/claude-plugin",
-  "version": "3.1.17",
+  "version": "3.1.19",
   "private": false,
   "description": "Claude Code plugin adapter for Webpresso agent-kit skills, commands, and MCP runtime.",
   "homepage": "https://github.com/webpresso/agent-kit#readme",

package/plugin-skill-ownership.json

@@ -2,9 +2,12 @@
   "schemaVersion": 1,
   "host": "claude",
   "packageName": "@webpresso/claude-plugin",
-  "packageVersion": "3.1.17",
+  "packageVersion": "3.1.19",
   "runtimeDirs": [".claude/skills"],
   "skills": {
+    "ai-deslop": {
+      "digest": "sha256:c1da1003009073dd3923454947e1141efc2e7df5ad8d725a511ed5cff03bfb5c"
+    },
     "autopilot": {
       "digest": "sha256:9e0b47c8438fe7c8123bec3361fcb17fd2fcd818a840f46ea272a7712e4c0b10"
     },
@@ -21,7 +24,7 @@
       "digest": "sha256:2af483ef80c5e0cf3d6d3da94d54b5bd1e9cecc9b2bd05e085499c2745c2c238"
     },
     "codex": {
-      "digest": "sha256:e8289cb51d023d8b20c1bfb2e349ee89c4d04cdfd5a03c7681328b3e60aab554"
+      "digest": "sha256:98187c2845b5cd88ba9eb9f682d16fec648932a084b3d1cebd066945b1fb6fdf"
     },
     "deep-research": {
       "digest": "sha256:01dafd194066a2936db0b437fd5c3beef6d7ff82478d0d995393093d89496f18"
@@ -33,11 +36,14 @@
       "digest": "sha256:306355888da0827e3523022077f731845726bbaaf595ca12515cc91d8605b5e6"
     },
     "fix": {
-      "digest": "sha256:e43dfad198f7279ebc2dca9e62f43fd960c783ca0ab25f133668fca13aed96b2"
+      "digest": "sha256:647aa94890e06b38b08e6cc12cd219343e963b4d3b469c349610b3a33f559dd5"
     },
     "hooks-doctor": {
       "digest": "sha256:3bfb0af64ff6796c56d51e4cd2bede7fb24c869f89b292546ea13ad2f5d3bb73"
     },
+    "investigate": {
+      "digest": "sha256:2051bccb2daf97ee91a8c15c7ef5bc7cc705d49beb4cbc7c8f6c63f4acda88a2"
+    },
     "lore-protocol": {
       "digest": "sha256:a80fbacc765e8b886437b9fdf7dd71e1ceb1b98b939078ec90fc88d06a8b4c69"
     },
@@ -57,7 +63,7 @@
       "digest": "sha256:b728dad90254d4c9f81b3818321d34ab53519b79ce20ff3bfe08bfacb2aa3e86"
     },
     "plan-refine": {
-      "digest": "sha256:891e84ff8592b8489b59e8642f6a602c9558fd57edd171af98499c94e2f62f13"
+      "digest": "sha256:1ad9175b999e946482109138ea23d4496f95987bf66ff41cf32fac5a764339a5"
     },
     "ralplan": {
       "digest": "sha256:db285d477f43ed23df990d4c03750d9916cdc7809356c1bf025efdf6a9add9ed"
@@ -78,7 +84,7 @@
       "digest": "sha256:dc6d3b2bd48a494d7a5cd30ad9c6c9999989b9920ca06ccd1f86a7cfaff60d54"
     },
     "verify": {
-      "digest": "sha256:ce0dd3a77508d1bf11c805dec2de04e75b689201d376b2ff0fa957604d584500"
+      "digest": "sha256:2964a54ef0f9db389d3ef5fa779fa2f095c9f8df47f19f3e4947d2403cdbf156"
     }
   }
 }

package/skills/ai-deslop/SKILL.md

@@ -0,0 +1,91 @@
+---
+type: skill
+slug: ai-deslop
+title: AI Deslop
+status: active
+scope: repo
+applies_to: [agents]
+related: [fix, verify, investigate]
+created: "2026-07-16"
+last_reviewed: "2026-07-16"
+name: ai-deslop
+description: "Bounded anti-slop cleanup pass: simplify, delete noise, or review-only with --review."
+license: MIT
+upstream:
+  source: https://github.com/Yeachan-Heo/oh-my-claudecode/tree/590fb988931d34a12604be0ca4215c818079018e/skills/ai-slop-cleaner
+  last_synced: "2026-07-16"
+argument-hint: "[--review] <files|path|scope>"
+---
+
+# AI Deslop
+
+Use this skill after behavior is already correct but the resulting code or docs
+still feel bloated, repetitive, over-scaffolded, or obviously machine-shaped.
+
+## Boundary
+
+- `ai-deslop` is a cleanup pass, not a feature lane.
+- Preserve behavior. If the work needs a logic change, route back to `fix`.
+- Prefer deletion and simplification over new abstraction.
+- Keep scope bounded to the requested files, changed files, or clearly named
+  area. Do not expand into broad redesign.
+
+## When to use it
+
+- The user explicitly asks to deslop, clean up AI slop, simplify, or trim noise.
+- A completed feature still contains dead code, duplicate helpers, wrapper
+  layers, weak boundaries, or generated-sounding prose/comments.
+- `verify` finds stale compatibility branches, cargo-cult abstractions, or
+  cleanup debt after the behavioral proof is already green.
+- The user wants a reviewer-only anti-slop pass via `--review`.
+
+## When not to use it
+
+- Behavior is still failing or root cause is not established.
+- The request is really a new feature, redesign, or semantic refactor.
+- Verification is too weak to protect preserved behavior.
+
+## Modes
+
+### Standard mode
+
+1. Lock behavior first with the narrowest regression proof available.
+2. Write a short cleanup plan naming the concrete slop to remove.
+3. Classify each smell:
+   - dead code / unused exports
+   - duplication / needless wrappers
+   - compatibility scaffolding that no longer serves a consumer
+   - noisy comments, docs, or prompts that restate the obvious
+   - weak boundaries or helper proliferation with no clear owner
+4. Apply the smallest safe deletions and consolidations first.
+5. Re-run the same scoped verification after each meaningful pass.
+
+### Review mode (`--review`)
+
+Use this to review a drafted cleanup pass without writing code.
+
+1. Inspect the cleanup plan, changed files, and verification evidence.
+2. Look specifically for leftover dead code, duplicate logic, needless
+   abstraction, and accidental behavior drift.
+3. Return a verdict plus required follow-ups.
+4. Do not both write and self-approve the same high-impact cleanup pass.
+
+## Verification
+
+Run only the narrowest checks that protect preserved behavior:
+
+- targeted tests
+- targeted lint/typecheck
+- any changed docs or command-surface checks
+
+If the cleanup touches shared/public surfaces or broad repo-owned docs, hand the
+result to `verify` before calling it done.
+
+## Output
+
+Return:
+
+- cleanup scope
+- slop classes removed
+- proof that behavior stayed locked
+- remaining follow-ups, if any

package/skills/codex/SKILL.md

@@ -50,6 +50,24 @@ fi
 
 Codex native subagent events are not currently parent-linked in the managed lifecycle. They are shown as unsupported, never inferred from text, PIDs, or timestamps.
 
+## Options
+
+Every mode runs through the `wp review run` runtime and honors these environment
+overrides (defaults shown). Set them inline, e.g.
+`CODEX_REVIEW_MODEL=gpt-5.6-sol CODEX_REVIEW_EFFORT=high /codex`.
+
+- `CODEX_REVIEW_MODEL` — the Codex model. Defaults to the `model = "..."` in the
+  active Codex config (`${CODEX_HOME:-~/.codex}/config.toml`), falling back to
+  `gpt-5.4`. It MUST be a model the current login is provisioned for: ChatGPT-account
+  logins use the config's codenamed ids (e.g. `gpt-5.6-sol`) and reject a bare
+  `gpt-5.6`. When the user names a model ("review with 5.6-sol"), pass it here.
+- `CODEX_REVIEW_EFFORT` — reasoning effort, default `medium`. The probe and
+  review stages accept only `medium` or `high`; any other value is rejected.
+- `CODEX_REVIEW_IDLE_SECONDS` / `CODEX_REVIEW_PROBE_IDLE_SECONDS` — idle timeout
+  (seconds) for the review stage and the preflight probe, default `180` each.
+- `CODEX_REVIEW_ARTIFACT_ROOT` — directory for the recorded review transcript
+  used as outside-review evidence, default `<cwd>/.webpresso/reviews`.
+
 ## Modes
 
 ### Review
@@ -59,6 +77,8 @@ Codex native subagent events are not currently parent-linked in the managed life
 3. Run Codex through the typed review runtime in read-only mode:
 
 ```bash
+# CODEX_REVIEW_MODEL / CODEX_REVIEW_EFFORT are overridable — see ## Options for
+# the full option list, valid values, and how to pick a login-supported model.
 CODEX_REVIEW_MODEL=${CODEX_REVIEW_MODEL:-gpt-5.4}
 CODEX_REVIEW_EFFORT=${CODEX_REVIEW_EFFORT:-medium}
 case "$CODEX_REVIEW_EFFORT" in

package/skills/fix/SKILL.md

@@ -9,7 +9,7 @@ related: [verify, testing-philosophy]
 created: "2026-05-13"
 last_reviewed: "2026-05-22"
 name: fix
-description: "Root-cause fix workflow for bugs, failures, regressions, and errors; reproduce, repair owner, test, verify."
+description: "Root-cause fix workflow that runs investigate first when diagnosis or failing proof is still missing."
 argument-hint: '<target> where target is: file|symptom|error|test|"free-text description"'
 ---
 
@@ -19,16 +19,9 @@ Fix the issue at its root, future-proof, aligned with repo philosophy.
 
 **Arguments**: $ARGUMENTS
 
-This command turns "please fix it well" into an enforceable workflow: reproduce, identify the broken invariant, repair it at the owner, prove the repair, and escalate to `/verify` when the blast radius is broader than a local fix.
-
-## Merged debugging discipline
-
-`/investigate` and `systematic-debugging` are folded into this workflow. Keep the
-old iron law: no fixes before evidence. Reproduce or capture the failure, read
-the complete error/stack, trace the bad value or broken invariant to its owner,
-compare with a working local pattern, and disconfirm cheap hypotheses before
-editing. If the issue crosses component boundaries, add temporary diagnostics at
-each boundary, run once, then remove or convert them to durable tests.
+This command turns "please fix it well" into an enforceable workflow: diagnose
+first, repair at the owner, prove the repair, and escalate to `/verify` when
+the blast radius is broader than a local fix.
 
 ## Non-negotiable invariants
 
@@ -42,6 +35,26 @@ each boundary, run once, then remove or convert them to durable tests.
 
 ## Protocol
 
+### Step 0 — Run the investigate phase when evidence is missing
+
+Before editing, check whether the current context already contains all of the
+following:
+
+- a named root-cause hypothesis
+- a failing proof tied to that hypothesis
+- the owning boundary that should be repaired
+
+If any of those are missing, run the `investigate` phase inline first:
+
+- reproduce or capture the failure
+- read the full error site and one nearby working pattern
+- disconfirm cheap alternative explanations
+- produce failing proof
+- stop only when you can name the broken invariant and owner
+
+Do not skip this phase because the fix "looks obvious". If the evidence is not
+there, the investigation work is part of `fix`.
+
 ### Step 1 — Reproduce and name the broken invariant
 
 - Run the exact command, test, or user flow that surfaces the failure.
@@ -98,6 +111,9 @@ Escalate to `/verify <target>` when any of these are true:
 - the fix touched docs, plans, blueprints, or repo SSOT
 - the fix needs dead-code / compat / broad regression review before claiming done
 
+If the repaired behavior is already green but the diff still needs a bounded
+cleanup pass, use `ai-deslop` before the final verify handoff.
+
 ### Step 6 — Report with evidence
 
 Keep it short:

package/skills/investigate/SKILL.md

@@ -0,0 +1,61 @@
+---
+name: investigate
+description: "Root-cause investigation workflow: reproduce, isolate, prove, and hand off the minimal repair scope."
+license: MIT
+---
+
+# Investigate
+
+Use this skill when the task is to understand a bug, failure, regression,
+flaky behavior, or confusing symptom before editing code.
+
+## Boundary
+
+- `investigate` owns diagnosis, reproduction, hypothesis testing, and proof.
+- `fix` owns repair. If the user asked to fix the problem and root-cause
+  evidence is still missing, `fix` should run this phase inline first.
+- Stop once you can name the broken invariant, show the failing proof, and
+  point at the owner that should be repaired.
+
+## Workflow
+
+1. Reproduce or capture the failure.
+   - Run the exact failing command or trace the exact user path.
+   - Keep the evidence that matters: error text, failing assertion, exit code,
+     log path, or boundary mismatch.
+2. Read the whole boundary.
+   - Inspect the full error site, the owning module, and one nearby working
+     pattern.
+   - Do not edit code yet.
+3. State the root-cause hypothesis.
+   - Name the broken invariant in one sentence.
+   - Name the owner that should enforce it.
+   - Separate facts from guesses.
+4. Disconfirm cheap alternatives.
+   - Test the obvious competing explanations before settling on one cause.
+   - Add temporary diagnostics only when needed, and remove them after one pass.
+5. Produce failing proof.
+   - Prefer a targeted failing test.
+   - If test-first is impossible at this boundary, use the nearest deterministic
+     reproduction instead.
+6. Hand off the minimal repair scope.
+   - List the owner, direct consumers, and the smallest safe verification set.
+   - If the issue is broader than one coherent repair, say so before any edit.
+
+## Output
+
+Return:
+
+- reproduced symptom
+- broken invariant
+- root-cause hypothesis status: confirmed / likely / blocked
+- failing proof
+- owning file or boundary
+- recommended next step: `fix`, `verify`, or a scoped blocker
+
+## Anti-patterns
+
+- Editing before you can name the invariant
+- Treating a stack trace as proof of root cause
+- Hiding uncertainty behind broad "probably X" explanations
+- Expanding a debugging pass into cleanup or refactor work

package/skills/plan-refine/SKILL.md

@@ -32,6 +32,7 @@ A plan is only as strong as its weakest unchecked assumption and only as fast as
 ## Mandatory gates
 
 - Apply DRY, SOLID, YAGNI, and KISS; reject speculative abstractions.
+- Choose the most elegant and durable solution the constraints allow, not merely a correct one: prefer deleting/reusing over adding, shrink public/export surface to the minimum real consumers require, keep the change scoped to one coherent idea, and reject the heavier option even when it works — record in the plan why the chosen shape is the durable one.
 - For package, catalog, generated-surface, or release changes, require package-surface and public-content checks.
 - Prefer deletion/reuse over new layers or dependencies.
 - Stop on unverified external claims, unresolved shared-file conflicts, or missing authority for destructive work.

package/skills/verify/SKILL.md

@@ -17,7 +17,9 @@ argument-hint: "<target> [--full] where target is: package|file|plan-slug|all"
 
 Post-implementation quality gate. Run after implementation exists, before claiming `done`.
 
-`/fix` owns reproduction, root-cause analysis, and the minimal repair. `/verify` owns the broader question: **is this work actually complete, integrated, and free of stale garbage?**
+`/fix` owns investigation-if-missing, root-cause analysis, and the minimal
+repair. `/verify` owns the broader question: **is this work actually complete,
+integrated, and free of stale garbage?**
 
 ## Iron law — evidence before claims
 
@@ -125,6 +127,10 @@ For command-surface, install/setup/update, or operator-guidance changes, explici
 
 Hard stop if a shared contract changed and consumers were not updated.
 
+Every session that changes behavior, commands, setup, or operator guidance must
+update every affected doc/help/instruction surface in the same change. Keep
+those doc edits concise and direct.
+
 ## Phase 4 — Test quality and behavioral audit
 
 The passing test suite is necessary, not sufficient.
@@ -149,6 +155,10 @@ Before claiming done:
 - justify every new exported/helper surface with a real consumer or test
 - run the repo's dead-code / dependency checks when the change broadened public surface or added new files
 
+If behavior is already locked and the remaining work is purely anti-slop
+cleanup, route that pass through `ai-deslop` first rather than broadening
+`verify` into a writer lane.
+
 This phase exists to catch the garbage that often slips in beside otherwise-correct work.
 
 ## Phase 6 — Outside-model approval gate
@@ -158,7 +168,7 @@ user explicitly asked for a different approval count or reviewer mix.
 
 Default policy:
 
-- Require **two extra model approvals** by default, but at most **one OpenCode Go reviewer** is required unless the user explicitly asks for multiple OpenCode models.
+- Require **one extra model approval** by default. At most **one OpenCode Go reviewer** counts toward the requirement unless the user explicitly asks for multiple OpenCode models.
 - If the user asks for more, fewer, zero, or specific reviewers, follow that
   explicit instruction and report the chosen requirement.
 - These approvals are model outside-voice approvals, not a substitute for
@@ -166,8 +176,8 @@ Default policy:
 
 Reviewer preference:
 
-- When running from Codex, prefer **Claude + one OpenCode Go reviewer**.
-- When running from Claude, prefer **Codex + one OpenCode Go reviewer**.
+- When running from Codex, prefer **Claude** (fall back to one OpenCode Go reviewer if Claude is unavailable).
+- When running from Claude, prefer **Codex** (fall back to one OpenCode Go reviewer if Codex is unavailable).
 - When Codex/Claude outside voice is unavailable, use **one OpenCode Go
   reviewer** if OpenCode is logged in.
 - Prefer the repo's outside-voice skills instead of hand-rolled commands: