@juancr11/sibu 0.5.0 → 0.5.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juancr11/sibu",
-  "version": "0.5.0",
+  "version": "0.5.2",
   "description": "CLI for setting up a local AI-augmented development workflow.",
   "repository": {
     "type": "git",
@@ -36,7 +36,8 @@
     "validate:packed-runtime": "node ./scripts/validate-packed-cli-runtime.mjs",
     "validate:doctor-version-advisory": "pnpm build && node ./scripts/validate-doctor-version-advisory.mjs",
     "validate:post-update-doctor-drift": "pnpm build && node ./scripts/validate-post-update-doctor-drift.mjs",
-    "validate:release": "pnpm verify && npm pack && pnpm run validate:packed-runtime && pnpm run validate:doctor-version-advisory && pnpm run validate:post-update-doctor-drift",
+    "validate:release-publish": "pnpm verify && npm pack && pnpm run validate:packed-runtime",
+    "validate:release": "pnpm run validate:release-publish && pnpm run validate:doctor-version-advisory && pnpm run validate:post-update-doctor-drift",
     "admin:release": "node ./bin/admin/release.js"
   },
   "keywords": [

package/templates/AGENTS.md

@@ -19,6 +19,14 @@
 - Ask again only if the scope changes materially, the approach becomes materially more complex or risky, or the user explicitly asks to review before continuing.
 - Use Conventional Commits 1.0.0 for commit messages.
 
+## Context budget discipline
+
+Treat context as a shared budget owned by the user. Prefer narrow, purposeful context before broad reads: search targeted paths, inspect snippets before full files, and avoid dependency, generated, build, cache, and lockfile content unless relevant.
+
+Do not dump full files, full diffs, broad recursive scans, or uncapped command output unless needed for quality or explicitly requested. Summarize large diffs, logs, command output, and generated artifacts by default, with focused excerpts when they support a decision.
+
+Keep responses concise by default, but spend the context needed for correctness, safety, validation, human control, or required context gathering. Warn or ask before optional expensive context operations.
+
 ## Communication style
 
 - Keep responses as short as practical while still being clear and useful.

package/templates/manifest.json

@@ -1,11 +1,11 @@
 {
-  "templateVersion": "49",
+  "templateVersion": "54",
   "templates": {
     "AGENTS.md": {
-      "version": "21",
+      "version": "22",
       "description": "Project-level agent instructions and Sibu maintenance guidance.",
       "changes": [
-        "Tightens Sibu maintenance outcome guidance while preserving startup health checks, safety, confirmation, routing, and workflow repair instructions."
+        "Adds concise context-budget discipline so agents prefer narrow reads, summarize large outputs, and preserve quality over token savings."
       ]
     },
     ".codex/config.toml": {
@@ -52,10 +52,10 @@
       ]
     },
     "skills/technical-design-writer/SKILL.md": {
-      "version": "10",
+      "version": "11",
       "description": "Mandatory technical design writer skill installed once at the shared .agents/skills workspace path.",
       "changes": [
-        "Compresses technical-design wording while preserving required grounding, UX binding, delegation boundaries, output location, and smallest-useful-design guidance."
+        "Tightens final responses so technical design updates report only the output path unless inline review is requested."
       ]
     },
     "skills/typescript/SKILL.md": {
@@ -115,10 +115,10 @@
       ]
     },
     "skills/ai-implementation-plan-executor/SKILL.md": {
-      "version": "8",
+      "version": "12",
       "description": "Mandatory AI implementation plan executor skill for implementing existing story implementation plans one reviewed step at a time.",
       "changes": [
-        "Adds context reuse guidance so story implementation plans load broad context once, then continue step-by-step using cached context unless scope, validation, or file changes require rereading."
+        "Clarifies focused approved-step commits and validation-only step reporting."
       ]
     },
     "skills/ai-prompt-engineer-master/SKILL.md": {
@@ -129,10 +129,10 @@
       ]
     },
     "skills/ux-expert/SKILL.md": {
-      "version": "2",
+      "version": "3",
       "description": "Selectable UX expert skill for UI-changing features, responsive design, flows, states, accessibility, and binding mockups.",
       "changes": [
-        "Compresses UX guidance while preserving product grounding, no-code boundary, phone-first design, output location, concrete mockup requirements, and binding downstream mockup authority."
+        "Tightens final responses so UX spec updates report only the output path unless inline review is requested."
       ]
     }
   }

package/templates/skills/ai-implementation-plan-executor/SKILL.md

@@ -119,7 +119,11 @@ When the user explicitly approves the current step, update that step file by add
 - Approved at: <ISO-8601 timestamp>
 ```
 
-Before writing the approval marker, identify the current Git user with `git config user.name`; if it is unavailable, use `git config user.email`. Use that value for `Approved by`. After writing the approval marker, commit all changes for the approved step before continuing. Use a Conventional Commits 1.0.0 message that describes the completed step. If the commit fails, stop and report the failure instead of continuing.
+Before writing the approval marker, identify the current Git user with `git config user.name`; if it is unavailable, use `git config user.email`. Use that value for `Approved by`.
+
+After writing the approval marker, commit only the changes produced by the approved step before continuing. The commit must include the step file approval marker and files intentionally changed while implementing that step. It must not include unrelated local edits, pre-existing worktree changes, or changes from other steps. Use the files tracked during step execution and a focused `git status --short` check to stage the correct paths. Do not run broad `git diff` investigations or other "what changed?" archaeology unless it is required to avoid committing unrelated changes and the user has approved that extra inspection.
+
+Use a Conventional Commits 1.0.0 message that describes the completed step. If the commit fails, stop and report the failure instead of continuing.
 
 Then continue with the next unapproved step immediately, without asking for another pre-implementation confirmation, only after the approval marker is written and the approved step changes are committed.
 
@@ -161,10 +165,10 @@ Do not:
 
 After implementing one step, briefly report:
 
+- that the step finished
+- if the step was validation-only and produced no code changes, say that it was just a validation step and all validations passed
 - current step file path
-- summary of changes made
-- validation run and result
-- any risks, blockers, or follow-up questions
+- the next step file path, or that no next step remains
 - that you are waiting for user approval before marking the step approved, committing it, and continuing to the next step
 
 After approving and committing the final step in a story implementation plan, also briefly report the Epic continuation check result:

package/templates/skills/technical-design-writer/SKILL.md

@@ -110,3 +110,7 @@ Use this structure as a starting point. Delete sections that do not add value.
 ## Quality bar
 
 A good technical design is short, specific, and useful. It should not try to be the product brief, architecture skill, clean-code skill, implementation plan, or ticket backlog.
+
+## Final response behavior
+
+After writing the file, report only the path. Do not paste the full technical design unless the user asks to review it inline.

package/templates/skills/ux-expert/SKILL.md

@@ -89,3 +89,7 @@ Use only helpful sections from this shape:
 ```
 
 The Binding Mockups section is authoritative for downstream work unless this UX spec is revised.
+
+## Final response behavior
+
+After writing the file, report only the path. Do not paste the full UX spec unless the user asks to review it inline.