@kody-ade/kody-engine-lite 0.1.49 → 0.1.51

package/README.md

@@ -13,15 +13,17 @@ Most AI coding tools are **autocomplete** (Copilot) or **chat-based** (Cursor, C
 
 | | Kody | Copilot Workspace | Devin | Cursor Agent |
 |---|---|---|---|---|
+| **Repo-aware instructions** | AI-customized per-stage prompts with your repo's patterns, gaps, and acceptance criteria | Generic prompts | Generic prompts | Generic prompts |
 | **Runs in CI** | GitHub Actions | GitHub Cloud | Devin Cloud | Local IDE |
 | **Fire and forget** | Yes | No — interactive | Partially | No — IDE must be open |
 | **Pipeline stages** | 7 stages with quality gates | Plan → implement | Single agent | Single agent |
 | **Shared sessions** | Stages share Claude Code sessions (no cold starts) | Single conversation | Single conversation | Single conversation |
 | **Risk gate** | Pauses HIGH-risk for human approval | No | No | No |
 | **AI failure diagnosis** | Classifies errors before retry (fixable/infra/abort) | No | No | No |
+| **Incremental codebase improvement** | Step files encode gaps to fix — every task raises quality | No | No | No |
 | **Model flexible** | Any LLM via LiteLLM | GitHub models only | Proprietary | Cursor models |
 | **Open source** | MIT | Proprietary | Proprietary | Proprietary |
-| **Cost** | Your API costs only | $10-39/month | $20-500/month | Subscription |
+| **Cost** | **Free** with free-tier models (Gemini, etc.) or pay-per-use with any LLM | $10-39/month | $20-500/month | Subscription |
 
 [Full comparison →](docs/COMPARISON.md)
 
@@ -117,7 +119,14 @@ cd your-project
 kody-engine-lite init
 ```
 
-This analyzes your project and generates workflow, config, memory, and labels — then commits and pushes.
+This analyzes your project and generates:
+- **Workflow** (`.github/workflows/kody.yml`)
+- **Config** (`kody.config.json` — auto-detected quality commands, git, GitHub settings)
+- **Project memory** (`.kody/memory/` — architecture and conventions)
+- **Customized step files** (`.kody/steps/` — see below)
+- **GitHub labels** for lifecycle tracking
+
+Then commits and pushes everything.
 
 ### 4. Use
 
@@ -172,6 +181,60 @@ kody-engine-lite init [--force]
 
 ## Key Features
 
+### Repo-Aware Step Files (`.kody/steps/`)
+
+**This is what separates Kody from every other AI coding tool.**
+
+Most AI tools send the same generic prompt regardless of whether they're working on a Next.js app, a Go microservice, or a Python ML pipeline. Kody generates **customized instruction files for every pipeline stage**, tailored to your specific repository.
+
+During `kody init`, Kody analyzes your codebase — your frameworks, patterns, conventions, file structure, even your anti-patterns — and produces 6 step files in `.kody/steps/`:
+
+| Step File | What It Controls |
+|-----------|-----------------|
+| `taskify.md` | How tasks are classified and scoped for this repo |
+| `plan.md` | Planning guidelines with your repo's architecture in mind |
+| `build.md` | Coding instructions with your actual patterns as examples |
+| `autofix.md` | How to fix verification failures using your toolchain |
+| `review.md` | Review checklist calibrated to your repo's quality bar |
+| `review-fix.md` | How to address review findings in your codebase |
+
+Each step file contains three repo-specific sections:
+- **Repo Patterns** — real code examples from your codebase (file paths, function signatures, snippets) showing what "good" looks like in your project
+- **Improvement Areas** — gaps, anti-patterns, and inconsistencies the AI should fix when it touches related code (e.g., "collections missing access control", "services not using DI")
+- **Acceptance Criteria** — a concrete checklist that defines "done" for each stage in your specific repo
+
+**Why this matters:**
+- Code produced by Kody **matches your existing patterns** — same naming, same abstractions, same file organization
+- Known gaps get **incrementally fixed** — every task that touches a weak area leaves it better
+- Quality bar is **explicit and auditable** — acceptance criteria are version-controlled in your repo, not hidden in a prompt
+- **You own and customize** the instructions — edit `.kody/steps/build.md` to change how Kody writes code, no engine changes needed
+- Re-run `kody init --force` after major refactors to **regenerate** step files from current state
+
+```
+# Example: .kody/steps/build.md (auto-generated, then customizable)
+---
+name: build
+...
+---
+<original engine prompt preserved>
+
+## Repo Patterns
+**Collection pattern** (`src/collections/certificates.ts`):
+- Always cast `relationTo` with `as CollectionSlug`
+- Register every new collection in `src/payload.config.ts`
+
+## Improvement Areas
+- Collections missing access control — add `access` guards when touching
+- Run `pnpm payload generate:types` after any schema change
+
+## Acceptance Criteria
+- [ ] `pnpm tsc --noEmit` exits with zero errors
+- [ ] All new collections include explicit access control
+- [ ] User-supplied strings sanitized via `src/security/sanitizers.ts`
+```
+
+### Other Features
+
 - **Shared Sessions** — stages in the same group share a Claude Code session, eliminating cold-start codebase re-exploration ([details](docs/FEATURES.md#shared-sessions))
 - **Risk Gate** — HIGH-risk tasks pause for human plan approval before building ([details](docs/FEATURES.md#risk-gate))
 - **AI Failure Diagnosis** — classifies errors as fixable/infrastructure/pre-existing/abort before retry ([details](docs/FEATURES.md#ai-powered-failure-diagnosis))

package/dist/bin/cli.js

@@ -1651,9 +1651,14 @@ function executeShipStage(ctx, _def) {
     const existingPr = getPRForBranch(head);
     if (existingPr) {
       updatePR(existingPr.number, body);
-      if (ctx.input.issueNumber && !ctx.input.local) {
+      if (!ctx.input.local) {
+        const msg = `\u2705 Fix pushed to PR #${existingPr.number}: ${existingPr.url}`;
         try {
-          postComment(ctx.input.issueNumber, `\u2705 Fix pushed to existing PR: ${existingPr.url}`);
+          if (ctx.input.prNumber) {
+            postPRComment(ctx.input.prNumber, msg);
+          } else if (ctx.input.issueNumber) {
+            postComment(ctx.input.issueNumber, msg);
+          }
         } catch {
         }
       }
@@ -2872,7 +2877,8 @@ async function main() {
     setGhCwd(projectDir);
     logger.info(`Working directory: ${projectDir}`);
   }
-  if (input.issueNumber && input.command !== "review") {
+  const isPRFix = input.command === "fix" && !!input.prNumber;
+  if (input.issueNumber && input.command !== "review" && !isPRFix) {
     const taskAction = resolveForIssue(input.issueNumber, projectDir);
     logger.info(`Task action: ${taskAction.action}`);
     if (taskAction.action === "already-completed") {
@@ -2904,7 +2910,9 @@ async function main() {
   }
   let taskId = input.taskId;
   if (!taskId) {
-    if (input.issueNumber) {
+    if (isPRFix) {
+      taskId = `fix-pr-${input.prNumber}-${generateTaskId()}`;
+    } else if (input.issueNumber) {
       taskId = `${input.issueNumber}-${generateTaskId()}`;
     } else if (input.command === "run" && input.task) {
       taskId = generateTaskId();
@@ -3005,7 +3013,17 @@ async function main() {
     fs20.writeFileSync(path19.join(taskDir, "task.md"), input.task);
   }
   const taskMdPath = path19.join(taskDir, "task.md");
-  if (!fs20.existsSync(taskMdPath) && input.issueNumber) {
+  if (!fs20.existsSync(taskMdPath) && isPRFix && input.prNumber) {
+    logger.info(`Fetching PR #${input.prNumber} details as task context...`);
+    const prDetails = getPRDetails(input.prNumber);
+    if (prDetails) {
+      const taskContent = `# ${prDetails.title}
+
+${prDetails.body ?? ""}`;
+      fs20.writeFileSync(taskMdPath, taskContent);
+      logger.info(`  Task loaded from PR #${input.prNumber}: ${prDetails.title}`);
+    }
+  } else if (!fs20.existsSync(taskMdPath) && input.issueNumber) {
     logger.info(`Fetching issue #${input.issueNumber} body as task...`);
     const issue = getIssue(input.issueNumber);
     if (issue) {
@@ -3092,6 +3110,7 @@ ${input.feedback}` : reviewContext;
       fromStage: input.fromStage,
       dryRun: input.dryRun,
       issueNumber: input.issueNumber,
+      prNumber: input.prNumber,
       feedback: input.feedback,
       local: input.local,
       complexity: input.complexity

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kody-ade/kody-engine-lite",
-  "version": "0.1.49",
+  "version": "0.1.51",
   "description": "Autonomous SDLC pipeline: Kody orchestration + Claude Code + LiteLLM",
   "license": "MIT",
   "type": "module",

package/templates/kody.yml

@@ -54,6 +54,7 @@ jobs:
       mode: ${{ steps.parse.outputs.mode }}
       from_stage: ${{ steps.parse.outputs.from_stage }}
       issue_number: ${{ steps.parse.outputs.issue_number }}
+      pr_number: ${{ steps.parse.outputs.pr_number }}
       feedback: ${{ steps.parse.outputs.feedback }}
       valid: ${{ steps.parse.outputs.valid }}
     steps:
@@ -108,7 +109,7 @@ jobs:
 
           # Validate mode
           case "$MODE" in
-            full|rerun|fix|status|approve) ;;
+            full|rerun|fix|status|approve|review) ;;
             *)
               # If first arg isn't a mode, it might be a task-id or nothing
               if [ -n "$MODE" ] && [ "$MODE" != "" ]; then
@@ -143,10 +144,22 @@ jobs:
             TASK_ID="${ISSUE_NUM}-$(date +%y%m%d-%H%M%S)"
           fi
 
+          # Detect if this comment is on a PR (PRs are issues in GitHub API)
+          PR_NUMBER=""
+          if [ -n "${{ github.event.issue.pull_request }}" ]; then
+            PR_NUMBER="$ISSUE_NUM"
+          fi
+
+          # For review mode on a PR comment: use the PR number directly
+          if [ "$MODE" = "review" ] && [ -n "$PR_NUMBER" ]; then
+            TASK_ID="review-pr-${PR_NUMBER}-$(date +%y%m%d-%H%M%S)"
+          fi
+
           echo "task_id=$TASK_ID" >> $GITHUB_OUTPUT
           echo "mode=$MODE" >> $GITHUB_OUTPUT
           echo "from_stage=$FROM_STAGE" >> $GITHUB_OUTPUT
           echo "issue_number=$ISSUE_NUM" >> $GITHUB_OUTPUT
+          echo "pr_number=$PR_NUMBER" >> $GITHUB_OUTPUT
           {
             echo "feedback<<KODY_EOF"
             echo "$FEEDBACK"
@@ -179,8 +192,8 @@ jobs:
           persist-credentials: true
           token: ${{ steps.app-token.outputs.token || secrets.GITHUB_TOKEN }}
 
-      - name: Checkout PR branch (for fix/rerun on PRs)
-        if: github.event.issue.pull_request && (needs.parse.outputs.mode == 'fix' || needs.parse.outputs.mode == 'rerun')
+      - name: Checkout PR branch (for fix/rerun/review on PRs)
+        if: github.event.issue.pull_request && (needs.parse.outputs.mode == 'fix' || needs.parse.outputs.mode == 'rerun' || needs.parse.outputs.mode == 'review')
         env:
           GH_TOKEN: ${{ steps.app-token.outputs.token || secrets.GITHUB_TOKEN }}
         run: |
@@ -219,6 +232,7 @@ jobs:
           MODE: ${{ github.event.inputs.mode || needs.parse.outputs.mode || 'full' }}
           FROM_STAGE: ${{ github.event.inputs.from_stage || needs.parse.outputs.from_stage }}
           ISSUE_NUMBER: ${{ github.event.inputs.issue_number || needs.parse.outputs.issue_number }}
+          PR_NUMBER: ${{ needs.parse.outputs.pr_number }}
           FEEDBACK: ${{ github.event.inputs.feedback || needs.parse.outputs.feedback }}
           DRY_RUN: ${{ github.event.inputs.dry_run || 'false' }}
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
@@ -226,8 +240,10 @@ jobs:
           CMD="run"
           [ "$MODE" = "rerun" ] && CMD="rerun"
           [ "$MODE" = "fix" ] && CMD="fix"
+          [ "$MODE" = "review" ] && CMD="review"
           ARGS="--issue-number $ISSUE_NUMBER"
           [ -n "$TASK_ID" ] && ARGS="$ARGS --task-id $TASK_ID"
+          [ -n "$PR_NUMBER" ] && ARGS="$ARGS --pr-number $PR_NUMBER"
           [ -n "$FROM_STAGE" ] && ARGS="$ARGS --from $FROM_STAGE"
           [ -n "$FEEDBACK" ] && ARGS="$ARGS --feedback \"$FEEDBACK\""
           [ "$DRY_RUN" = "true" ] && ARGS="$ARGS --dry-run"