feat(cc-sdd): skills migration with native subagent dispatch by gotalab · Pull Request #141 · gotalab/cc-sdd

gotalab · 2026-02-15T17:38:39Z

Summary

Inline migration: Remove context: fork from all 9 claude-code skills, add Parallel Research sections (both platforms)
Task decomposition: Add ordering principle (Foundation→Core→Integration→Validation), _Boundary_ and _Depends_ annotations
Intent clarification: RPT-inspired problem-first clarification for init and requirements skills
Next Phase guidance: Consistent workflow navigation across all claude-code skills
Ralph Loop support: New kiro-ralph-impl skill with Feature Flag Protocol for autonomous TDD implementation
AskUserQuestion: Added to allowed-tools for init, requirements, validate-design

Changed files (28)

11 claude-code skills (9 migrated + 1 new ralph-impl + 1 validate-design restructured)
9 codex skills (Parallel Research added)
3 shared rules/templates (tasks-generation, tasks-parallel-analysis, tasks template)
2 test files (assertions updated, 12→13 skills)
1 CLAUDE.md (workflow + skills description)

Test plan

npm test — 161/161 pass
No context: fork remaining in claude-code skills
Parallel Research in all 9 applicable skills (both platforms)
kiro-ralph-impl deployed with templates/ralph-prompt.md
Feature Flag Protocol in ralph-prompt.md

Note

Medium Risk
Medium risk because it adds/changes workflow-driving skill, template, and rules content that can alter generated specs and user guidance across installs, though it does not modify core runtime logic.

Overview
Adds a new skills package for integrating additional coding agents into cc-sdd. Introduces cc-sdd-new-agent (with SKILL.md, OpenAI agent metadata, and migration/planning reference docs) that standardizes a plan-first SOP for new agent additions and skills migrations.

Expands skills-mode process docs and templates. Adds a full set of .kiro/settings/rules/* and .kiro/settings/templates/specs/* documents covering discovery/design/task gates and task-parallelization conventions, plus updates top-level README and multiple guides to reposition cc-sdd around skills-mode entry points (/kiro-discovery, /kiro-spec-batch, /kiro-impl) and mark legacy command modes as deprecated.

Adds repo-level config for “entire” runs. Introduces .entire/settings.json and .entire/.gitignore for local execution artifacts/settings.

^{Reviewed by Cursor Bugbot for commit a8ee894. Bugbot is set up for automated code reviews on this repo. Configure here.}

…alph Loop support - Remove context:fork from all 9 claude-code skills, add Parallel Research sections - Add Parallel Research to all 9 codex skills - Add task ordering (Foundation→Core→Integration→Validation), _Boundary_ and _Depends_ annotations - Add RPT-inspired intent clarification to init and requirements skills - Add AskUserQuestion to allowed-tools for init, requirements, validate-design - Add Next Phase guidance to all claude-code skills - Add kiro-ralph-impl skill with Feature Flag Protocol for autonomous TDD implementation - Update tests (12→13 skills) and CLAUDE.md workflow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Add 4 loop management scripts to kiro-ralph-impl (setup-loop.sh, check-loop.sh, next-task.sh, complete-task.sh) adapted from CC plugin's ralph-loop - Update SKILL.md and ralph-prompt.md with script-driven orchestration protocol - Add agents/openai.yaml to all 13 skills per Codex skills spec (display_name, short_description, allow_implicit_invocation: false) - Add test assertions for scripts and openai.yaml existence Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…fication Ralph Loop scripts: - check-loop.sh: remove task counting, iteration guard only (CONTINUE/MAX_ITERATIONS_REACHED) - setup-loop.sh: filter subtasks only (X.Y pattern) for max_iterations calculation - next-task.sh: fix dep-blocked skip (continue not exit), fix _Blocked:/_Depends: pattern matching, add pre-task git snapshot - complete-task.sh: delta-based staging instead of git add -A - Update ralph-prompt.md and SKILL.md for both Claude Code and Codex Spec clarification: - init: clarify intent based on deliverable requirements (who/situation/change), not subjective ambiguity - requirements: clarify-then-generate instead of generate-then-iterate, add requirements vs design scope boundary (ask about behavior, defer technology to design) Design synthesis: - Add design-synthesis.md rule: generalization, build-vs-adopt, simplification - Insert Step 3 (Synthesis) between Discovery and Generate in design SKILL.md Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

… physical copies Replace physical rule file copies in skill directories with a declarative metadata.shared-rules field in SKILL.md frontmatter. cc-sdd now resolves rules from templates/shared/settings/rules/ at install time, eliminating agent×skill duplication (22 files removed, single source of truth maintained). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Default agent changed from claude-code to claude-code-skills so `npx cc-sdd` now installs 13 skills into .claude/skills/ by default - Fix skill count 12→13 in all READMEs (kiro-ralph-impl was added) - Add .claude/skills/ to project structure tree in READMEs - Clarify .kiro/settings/rules/ is for non-skills agents only - Update tests to match new default agent Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Highlight the SDD → Ralph Loop pipeline: spec-tasks generates tasks, ralph-impl executes them autonomously in a loop, making SDD specs self-executing plans. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Remove external ralph-loop plugin dependency and merge kiro-spec-impl + kiro-ralph-impl into a unified kiro-impl skill with dual mode: - Autonomous mode (no task args): dispatches fresh subagent per task with independent adversarial review - Manual mode (task numbers): TDD in main context Key changes: - New kiro-impl skill with implementer + reviewer prompt templates (superpowers-style, portable across agents) - Reviewer: 10-point checklist including reality check, spec alignment, regression safety, error handling, no hardcoded secrets - Subagent status protocol: DONE/DONE_WITH_CONCERNS/BLOCKED/NEEDS_CONTEXT - Remove agents_library manifest artifact (no more .claude/agents/) - Update all cross-references and docs (3 READMEs, agent docs, 6 skills) - 15 files deleted, 7 created, ~10 modified Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Show upgrade notice when users install with --claude or --claude-agent, recommending --claude-skills for subagent-driven autonomous implementation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Replace Ralph Loop messaging with native subagent-driven autonomous implementation across all READMEs (EN/JA/ZH-TW) and root README. Use "no external dependencies" instead of "no plugins" since Ralph Loop was a separate project, not bundled with cc-sdd. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ation Per-task checks (reality check, spec alignment, test quality) are now handled by the reviewer subagent during kiro-impl. Refocus kiro-validate-impl on feature-level concerns that only become visible across all tasks: - Cross-task integration (data flows, API contracts, shared state) - Requirements coverage gaps between task boundaries - Design end-to-end alignment (architecture as a whole) - Full test suite regression (not just per-task tests) - Blocked tasks impact assessment Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: aee5f59749ab

Before dispatching each task, the controller now synthesizes a concrete Task Brief from the abstract tasks.md entry by cross-referencing requirements.md and design.md. The brief includes: - Acceptance criteria (observable behaviors) - Completion definition (files, functions, tests) - Design constraints (technical decisions) - Verification method (how to confirm) - Behavioral classification (Feature Flag Protocol applicability) This gives implementer subagents superpowers-level specificity while keeping tasks.md kiro-compatible. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 3138a8104ce9

Task Brief construction now happens inside the implementer subagent instead of the controller. Each subagent reads only the spec sections it needs for its single task, keeping the controller lightweight. If the spec is too vague to derive concrete acceptance criteria, design constraints, or verification method, the implementer reports NEEDS_CONTEXT instead of guessing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 5d9b1bcd87d5

Prevent controller context exhaustion over 20+ task runs: - One sub-task per iteration (NEVER batch 3.1+3.2 into one dispatch) - Re-read tasks.md every iteration (don't rely on accumulated memory) - Discard full status reports after each iteration, keep one-line summary These constraints prevent the two failure modes: context overflow from accumulated details, and quality degradation from batching tasks. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 5aea33f93fe5

- Reviewer receives git diff as primary input (not just implementer self-report). "Do Not Trust the Report" is now enforceable. - (P) parallel markers are informational only; kiro-impl processes all tasks sequentially to avoid git conflicts. - Add session resume guidance: safe to re-run after interruption since completed tasks are [x] in tasks.md and committed to git. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: e8a14fac5725

Move git diff execution from controller to reviewer subagent, same pattern as Task Brief. Controller passes task description + spec paths, reviewer reads the actual changes and spec sections independently. Keeps controller lightweight and avoids bloating the dispatch prompt with potentially large diffs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: f5561e575633

Two process improvements inspired by superpowers analysis: 1. kiro-brainstorm (optional, both agents): - Refines vague ideas into concrete project descriptions - Structured dialogue: explore context → questions → 2-3 approaches → approve - No files created — conversation context carries forward to spec-init - Codebase-aware (reads steering, scans existing specs) 2. File Structure Plan in design.md: - New section in design template: New Files / Modified Files / Directory Layout - Each file must have one clear responsibility - kiro-spec-design now explicitly generates this section - Directly drives task _Boundary:_ annotations and Task Briefs Context engineering: brainstorm gathers codebase context silently before asking questions. Harness engineering: file structure plan creates the concrete anchor that flows through tasks → impl → review. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 3cc956cd14f8

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: f9948520e0d5

Rules are installed per-skill (e.g., .claude/skills/kiro-spec-requirements/rules/) not in .kiro/settings/rules/. Fixed paths so subagents can find EARS format, review gates, design principles, and task generation rules. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: dee375ebec1b

…atch Instead of listing every template/rule path in the subagent prompt, point the subagent to each skill's SKILL.md as the source of truth. The subagent reads the SKILL.md to discover templates, rules, and review gates. Reduces duplication and improves maintainability. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 43b6fbddcce8

- spec-quick: Replace SlashCommand with Skill tool, remove TodoWrite dependency, use simple text progress instead - spec-status: Replace $1 with $ARGUMENTS, trim allowed-tools to read-only (Read, Glob, Grep), remove legacy XML tags - validate-gap: Add Write to allowed-tools, persist gap analysis to research.md so it survives session boundaries - spec-design: Remove redundant design-principles.md re-read in Step 4 - spec-batch: Add context budget (~3000 lines) to cross-spec review subagent, prioritize design.md over tasks.md for review Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: b2e0b274598c

Subagent has fresh context, hard line limit unnecessary. Keep reading priority guidance (design.md > requirements.md > tasks.md boundaries). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: a06e5ed00dc5

1000 lines is a more realistic threshold for spec splitting. Reduces context pressure on downstream skills that read design.md in main context. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: ccfad91a7317

…cies High priority: - Add {{LANG_CODE}} placeholder to spec-init and spec-quick (both agents) - spec-design now reads research.md in Step 1 (gap analysis handoff) - brainstorm references correct approvals structure instead of non-existent approved field - spec-design Step 6 formalizes research.md write Medium: - Remove stale TodoWrite reference from spec-quick Phase 1 - Fix stale /kiro:spec-requirements syntax in requirements-init.md template - Fix incomplete -y comment in spec-quick Phase 4 - Fix overstated auto mode banner (internal review gates still run) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 25c1a13fcb76

- Remove model: "opus" hint from spec-batch (unsupported by Agent tool) - Codex spec-quick: remove all task list references, align with claude-code-skills version (text progress, clean phase invocations) - Fix Phase 3/4 output progress format in codex spec-quick Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: e5dc78fcfc8c

- validate-impl codex: add Mechanical first, Remediation required constraints, output items, and Session Interrupted fallback - spec-design claude-code: add research.md template loading, Persist Findings sub-step, and Research Log output item (from codex) - impl codex: add Test Failures error scenario - steering codex: fix stale tool names (glob_file_search → Glob, etc.) - steering-custom codex: add Step 1 Gather Context, fix LS tool name Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: f48fa46f5d93

Instruct design to derive test items from requirements' acceptance criteria, reference specific components, and map E2E paths to critical user flows. Prevents generic/vague test entries in design.md. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 100f40175864

If there is even a 1% chance a skill applies, invoke it. Prevents the model from skipping skills for tasks that seem simple. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: f53873540605

… gates 1. Implementer must capture failing test output (RED phase) and include RED_PHASE_OUTPUT in status report as evidence tests were written first 2. Reviewer checks for RED_PHASE_OUTPUT presence for behavioral tasks, rejects if missing 3. Requirements review gate: mechanical checks for numeric IDs, EARS acceptance criteria presence, implementation language detection 4. Design review gate: mechanical checks for requirements traceability, File Structure Plan population, orphan component detection Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 2522272b0361

… lines) Systematic removal of instructions that duplicate execution steps: - Role sections (restated frontmatter description): all 10 files - Mission lines (restated description): all 10 files - Tool Guidance sections (restated step-level tool usage): 8 files - Critical Constraints duplicating steps: 7 files - Output Description duplicating final steps: 5 files - Safety items duplicating step logic: 6 files Kept: Success Criteria, numeric constraints, structured output formats, error scenarios with specific messages, non-obvious constraints. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 6ab88b56ad19

…62 lines) Mirror the claude-code-skills trimming: remove Role/Mission duplication, Tool Guidance sections, redundant constraints, duplicate Safety items, and Output Description sections that repeat final steps. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: e0dd8341d6f8

READMEs (4 files): - Skill count 12/13 → 14 (kiro-spec-batch added) - --codex prompts mode: "legacy" → "blocked, use --codex-skills" - Version v2.0.0 → v3.0.0 - Add brainstorm entry point and spec-batch to use cases Guides (4 files): - spec-driven: brainstorm as entry point with action paths, spec-batch row, brief.md persistence, cross-spec review - command-reference: brainstorm/spec-batch in skill lists - claude-subagents: brainstorm entry point, spec-batch parallel creation - migration-guide: new v2.x → v3.0 migration section Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 3cd2486b5e3f

"Turn approved specs into long-running autonomous implementation. A minimal, adaptable SDD harness." Key messaging: - Specs you can run (control plane, not documents) - Long-running autonomous implementation (kiro-impl) - Scales to real products (brainstorm + spec-batch) - Customize once, adapt as models improve Updated: root README, tools/cc-sdd/README (EN/JA/zh-TW), package.json Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 6f32075547e2

npx cc-sdd (no flags) now installs claude-code-skills by default in non-interactive mode. Interactive mode still shows the selection prompt. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 516a5f099a02

Add Agent Skills support for Cursor, GitHub Copilot, OpenCode, Gemini CLI, Windsurf, and Antigravity (+ windsurf-skills). Each platform gets 14 skills with platform-specific transformations from codex-skills canonical source. - Registry: 7 new agent definitions with correct layout, commands, prefixes - Manifests: 7 new manifest JSON files - Templates: 7 platform template directories (14 skills × 7 = 98 SKILL.md files) - Tests: 7 integration test files (21 new tests, 187 total) - Windsurf uses @kiro- prefix (not /kiro-), sequential-only fallback - Antigravity uses sequential-only fallback (no sub-agent dispatch) - Gemini gets gemini-agents/spec-reviewer.md for cross-spec review - Legacy modes show prominent DEPRECATED warning with migration command - Recommended models updated (Gemini 3.1 Pro, Composer 2 for Cursor) - Docs updated: skills modes as primary, legacy modes as deprecated Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 3fdfd8b3fb03

When implementer reports BLOCKED or reviewer rejects after 2 rounds, kiro-impl now spawns a fresh debug subagent instead of immediately skipping the task. The debug subagent investigates root causes with web search in a clean context (no failed implementation history), then a new implementer retries with the fix plan. Max 2 debug rounds. - Add debugger-prompt.md template (identical across all 8 platforms) - Update kiro-impl SKILL.md: (b) BLOCKED → debug, (d) REJECTED×2 → debug - Add (g) debug subagent section with rollback and re-dispatch flow - Inject previous task learnings (Implementation Notes) into implementer prompt - Bounded: remediation ×2 + debug ×2 = max 6 steps before BLOCKED Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 10180cb87bc8

Update EN and JA documentation to reflect the new debug subagent flow and Implementation Notes injection in kiro-impl autonomous mode. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: de975e8aaa28

…kills - Add auto-debug, learnings propagation to long-running impl descriptions - Update Agent Skills platform list to all 8 platforms - Mark legacy modes as deprecated across EN/JA/zh-TW READMEs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: da37d732b12b

TDD with Feature Flag Protocol is a core feature that was missing from the long-running impl descriptions. Web search is an implementation detail of the debug subagent, not README-level information. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 5c7362e643c0

- JA/zh-TW: Add --*-skills flags to install sections, update agent tables - Migration guide EN: Add 8-platform skills, debug subagent, learnings rows - Migration guide JA: Add v2.x → v3.0 section with full platform list Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: c42468bed66b

…able Keep technical terms (skill names, command names, area labels) in English for consistency across all language versions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: fa869b702209

…d names Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 14d6d81afaaa

Entire-Checkpoint: b25c9335f73d

Entire-Checkpoint: c7347312ea66

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit a8ee894. Configure here.}

cursor · 2026-04-08T23:11:07Z

.entire/settings.json

+  "strategy": "manual-commit",
+  "enabled": true,
+  "telemetry": false
+}


Local tool configuration committed to repository

Low Severity

The .entire/ directory containing settings.json (with strategy, enabled, telemetry fields) and a .gitignore appears to be local configuration for the "Entire" tool (a code-context tool). This is unrelated to the cc-sdd skills migration and looks like it was accidentally included in the commit. Similar to .vscode/settings.json or .idea/ files, tool-specific workspace configuration typically belongs in a global gitignore rather than committed to the repository.

Additional Locations (1)

.entire/.gitignore#L1-L4

^{Reviewed by Cursor Bugbot for commit a8ee894. Configure here.}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a8ee894b07

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-08T23:15:46Z

tools/cc-sdd/src/plan/sharedRules.ts

+export const SHARED_RULES_DIR = 'templates/shared/settings/rules';
+
+export const parseSharedRules = (content: string): string[] => {
+  const fmMatch = content.match(/^---\n([\s\S]*?)\n---/);


Accept CRLF frontmatter when parsing shared-rules

The frontmatter matcher only accepts \n line endings, so parseSharedRules returns no rules for SKILL.md files with CRLF (common on Windows checkouts with core.autocrlf=true). In this commit, skill-local rules/ files were removed and rules are injected only via this parser, so a CRLF checkout installs skills without required rules/*.md files, breaking workflows that reference those rule documents.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-08T23:15:46Z

tools/cc-sdd/src/cli/agents.ts

+  if (current) return current;
+  if (!isInteractive()) return DEFAULT_AGENT;


Let non-interactive runs honor configured agent

This now forces claude-code-skills whenever no CLI agent flag is provided in non-interactive mode, which makes args.agent always set before mergeConfigAndArgs. That overrides any configured agent value passed via config/runtime integration, so scripted runs can silently install the wrong agent templates even though config precedence is expected to be CLI > config > defaults.

Useful? React with 👍 / 👎.

gotalab and others added 4 commits February 16, 2026 02:38

WIP: add Claude/Codex skills support and docs updates

1ac072c

Add Codex spec-quick skill and collaboration mode guidance

646569a

Fix Codex recommended model list for non-released gpt-5.3

4d4f3f8

gotalab changed the title ~~WIP: Add Claude/Codex skills support and docs updates~~ Skills inline migration, task decomposition, and Ralph Loop support Mar 2, 2026

gotalab and others added 10 commits March 3, 2026 15:25

Add Ralph Loop integration to What's New and Key Features in READMEs

c062f6e

Highlight the SDD → Ralph Loop pipeline: spec-tasks generates tasks, ralph-impl executes them autonomously in a loop, making SDD specs self-executing plans. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

feat(cc-sdd): harden Ralph Loop validation flows

2d40eb5

chore(cc-sdd): update recommended OpenAI models

63c0dbe

docs(cc-sdd): sharpen readme positioning

e391f66

chore(cc-sdd): update Claude model recommendations

b7eca3f

feat(cc-sdd): harden skill review and context loading

2809750

gotalab mentioned this pull request Mar 7, 2026

Auto context management on subagents mode #147

Open

gotalab added 2 commits March 9, 2026 03:28

feat(cc-sdd): harden requirements and final validation gates

ea29ede

chore(cc-sdd): relax gpt-5.4 reasoning recommendation

85cb6ba

takaram mentioned this pull request Apr 3, 2026

cc-sdd no longer works with Codex CLI #152

Open

gotalab changed the title ~~Skills inline migration, task decomposition, and Ralph Loop support~~ feat(cc-sdd): skills migration with native subagent dispatch Apr 5, 2026

gotalab and others added 9 commits April 5, 2026 21:57

gotalab and others added 27 commits April 6, 2026 02:58

chore(cc-sdd): increase cross-spec review remediation limit to 3 rounds

b7daa14

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: f9948520e0d5

feat(cc-sdd): add 1% skill invocation rule to docs

fd04b76

If there is even a 1% chance a skill applies, invoke it. Prevents the model from skipping skills for tasks that seem simple. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: f53873540605

docs(cc-sdd): revert JA migration guide table to Japanese with comman…

4c68c7a

…d names Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Entire-Checkpoint: 14d6d81afaaa

feat: align skills around boundary-first spec flow

e199d9e

Entire-Checkpoint: b25c9335f73d

feat: strengthen boundary-first spec workflow

a8ee894

Entire-Checkpoint: c7347312ea66

gotalab marked this pull request as ready for review April 8, 2026 23:10

cursor bot reviewed Apr 8, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Apr 8, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(cc-sdd): skills migration with native subagent dispatch#141

feat(cc-sdd): skills migration with native subagent dispatch#141
gotalab wants to merge 76 commits intomainfrom
wip/claude-codex-skills-support

gotalab commented Feb 15, 2026 •

edited by cursor bot

Loading

Uh oh!

cursor bot left a comment

Uh oh!

cursor bot Apr 8, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Apr 8, 2026

Uh oh!

chatgpt-codex-connector bot Apr 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		if (current) return current;
		if (!isInteractive()) return DEFAULT_AGENT;

Conversation

gotalab commented Feb 15, 2026 • edited by cursor bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changed files (28)

Test plan

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor bot Apr 8, 2026

Choose a reason for hiding this comment

Local tool configuration committed to repository

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

gotalab commented Feb 15, 2026 •

edited by cursor bot

Loading