@windyroad/architect 0.18.0-preview.802 → 0.18.1-preview.804

package/.claude-plugin/plugin.json

@@ -123,5 +123,5 @@
     }
   },
   "name": "wr-architect",
-  "version": "0.18.0"
+  "version": "0.18.1"
 }

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "wr-architect",
-  "version": "0.18.0",
+  "version": "0.18.1",
   "description": "Architecture decision enforcement for AI coding agents",
   "author": {
     "name": "Windy Road Technology",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@windyroad/architect",
-  "version": "0.18.0-preview.802",
+  "version": "0.18.1-preview.804",
   "description": "Architecture decision enforcement for AI coding agents",
   "bin": {
     "windyroad-architect": "./bin/install.mjs"
@@ -27,6 +27,7 @@
     ".claude-plugin/",
     ".codex-plugin/",
     "lib/",
-    "!skills/*/eval/"
+    "!skills/*/eval/",
+    "!agents/eval/"
   ]
 }

package/agents/eval/codex-rubric-output.schema.json

@@ -1,22 +0,0 @@
-{
-  "type": "object",
-  "additionalProperties": false,
-  "required": [
-    "pass",
-    "score",
-    "reason"
-  ],
-  "properties": {
-    "pass": {
-      "type": "boolean"
-    },
-    "score": {
-      "type": "number",
-      "minimum": 0,
-      "maximum": 1
-    },
-    "reason": {
-      "type": "string"
-    }
-  }
-}

package/agents/eval/grade-codex-rubric.sh

@@ -1,39 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-REPO_ROOT="$(cd "${SCRIPT_DIR}/../../../.." && pwd)"
-SCHEMA="${SCRIPT_DIR}/codex-rubric-output.schema.json"
-PROMPT="${*:-}"
-
-if [[ -z "$PROMPT" ]]; then
-  echo "grade-codex-rubric.sh: rubric prompt argument is required" >&2
-  exit 2
-fi
-
-raw="$(
-  codex exec \
-    --ephemeral \
-    --cd "$REPO_ROOT" \
-    -c 'approval_policy="never"' \
-    --sandbox read-only \
-    --output-schema "$SCHEMA" \
-    "You are a strict grading assistant. Respond only with JSON matching the provided schema. Grade this promptfoo rubric literally.
-
-${PROMPT}"
-)"
-
-printf '%s' "$raw" | awk '
-  BEGIN { depth = 0; started = 0 }
-  {
-    line = $0
-    for (i = 1; i <= length(line); i++) {
-      c = substr(line, i, 1)
-      if (c == "{") { depth++; started = 1 }
-      if (started) { buf = buf c }
-      if (c == "}") { depth--; if (depth == 0 && started) { print buf; exit } }
-    }
-    if (started) { buf = buf "\n" }
-  }
-  END { if (started && depth != 0) print buf }
-' || printf '%s' "$raw"

package/agents/eval/promptfooconfig.codex.yaml

@@ -1,56 +0,0 @@
-description: |
-  Codex wr-architect agent verdict eval. Exercises the generated
-  .codex/agents/wr-architect.toml path and the Codex plugin install surface.
-
-providers:
-  - id: 'exec:bash ./run-codex-agent-eval.sh'
-
-defaultTest:
-  options:
-    provider:
-      id: 'exec:bash ./grade-codex-rubric.sh'
-
-prompts:
-  - '{{prompt}}'
-
-tests:
-  - description: PASS for aligned pre-edit proposal
-    vars:
-      prompt: |
-        PRE-EDIT review: proposed change updates comments only. No
-        architecture-bearing files, dependencies, hooks, workflows, or ADRs
-        change. Return the architect verdict.
-    assert:
-      - type: regex
-        value: 'Architecture Review:\s*PASS|PASS'
-
-  - description: ISSUES FOUND for undocumented dependency
-    vars:
-      prompt: |
-        PRE-EDIT review: proposed change adds a new npm dependency named
-        fast-queue to package.json for core task scheduling. No existing ADR
-        covers this dependency choice. Return the architect verdict.
-    assert:
-      - type: regex
-        value: 'ISSUES FOUND|Undocumented Decision|new decision'
-
-  - description: NEEDS DIRECTION when viable options are unpinned
-    vars:
-      prompt: |
-        PRE-EDIT review: proposed change must choose whether plugin
-        distribution uses a remote marketplace only or also a repo-local
-        marketplace, and the prompt does not pin a direction. Return the
-        architect verdict.
-    assert:
-      - type: regex
-        value: 'NEEDS DIRECTION'
-
-  - description: Unratified Dependency flags unconfirmed ADR dependency
-    vars:
-      prompt: |
-        PRE-EDIT review: proposed change explicitly implements ADR-074, but
-        the ADR frontmatter lacks human-oversight: confirmed. Return the
-        architect verdict and action.
-    assert:
-      - type: regex
-        value: 'Unratified Dependency|ratify|review-decisions'

package/agents/eval/run-codex-agent-eval.sh

@@ -1,47 +0,0 @@
-#!/usr/bin/env bash
-# Codex promptfoo exec-provider driver for the generated wr-architect agent.
-set -euo pipefail
-
-SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-REPO_ROOT="$(cd "${SCRIPT_DIR}/../../../.." && pwd)"
-MARKETPLACE_DIR="$REPO_ROOT"
-PROMPT="${*:-}"
-
-if [[ -z "$PROMPT" ]]; then
-  echo "run-codex-agent-eval.sh: prompt argument is required" >&2
-  exit 2
-fi
-
-SOURCE_CODEX_HOME="${CODEX_HOME:-${HOME}/.codex}"
-TMP_CODEX_HOME=""
-
-if [[ -n "${WR_CODEX_EVAL_CODEX_HOME:-}" ]]; then
-  export CODEX_HOME="$WR_CODEX_EVAL_CODEX_HOME"
-  mkdir -p "$CODEX_HOME"
-else
-  TMP_CODEX_HOME="$(mktemp -d)"
-  chmod 700 "$TMP_CODEX_HOME"
-  export CODEX_HOME="$TMP_CODEX_HOME"
-  if [[ -f "${SOURCE_CODEX_HOME}/auth.json" ]]; then
-    cp "${SOURCE_CODEX_HOME}/auth.json" "$CODEX_HOME/auth.json"
-    chmod 600 "$CODEX_HOME/auth.json"
-  fi
-  if [[ "${WR_CODEX_EVAL_COPY_CONFIG:-0}" == "1" && -f "${SOURCE_CODEX_HOME}/config.toml" ]]; then
-    cp "${SOURCE_CODEX_HOME}/config.toml" "$CODEX_HOME/config.toml"
-    chmod 600 "$CODEX_HOME/config.toml"
-  fi
-  trap 'rm -rf "$TMP_CODEX_HOME"' EXIT
-fi
-
-codex plugin marketplace add "$MARKETPLACE_DIR" >/dev/null
-codex plugin add wr-architect@windyroad-local >/dev/null
-
-exec codex exec \
-  --ephemeral \
-  --cd "$REPO_ROOT" \
-  -c 'approval_policy="never"' \
-  --sandbox read-only \
-  --dangerously-bypass-hook-trust \
-  "Spawn the custom collaborator agent named wr-architect:agent (auto-discovered by Codex from the installed wr-architect plugin's agents/agent.md). If this non-interactive harness cannot spawn the custom agent directly, read .codex/agents/wr-architect.toml and perform the same review inline.
-
-${PROMPT}"