Planning with files

- Hooks now use a unified scripts/skill-hook.sh, replacing various script calls and event flags for simpler lifecycle handling. - Added templates/loop.md as a new planning template. - Removed the legacy skill-card.md file. - Updated documentation: task planning files are now resolved per selected task directory (with PLAN_ID) instead of always defaulting to project root. - Clarified setup and initialization steps for parallel or isolated task directories.

- PostToolUse progress reminder now reaches the model instead of the user, fires once per turn instead of once per tool call, and no longer triggers on read-only shell commands (issue #239). - An explicit PLAN_ID is now a binding: it resolves or nothing is injected, never a different plan (issue #237). - A project's root .mode is a floor a plan cannot start below, so creating a plan can no longer drop a committed attestation requirement (issue #238). - plan-doctor classifies injection output on the data framing, so a plan quoting a control phrase no longer reports a false tamper warning and a dark-hooks state no longer reports PASS (issue #236). - Version metadata updated to 3.16.0.

- Session recovery now reads only project planning files by default; local agent session history is accessed only via explicit user request with session-catchup.py. - Added support for session-catchup.py metadata mode (counts only, never emits transcript or commands) and bounded replay mode (non-default, opt-in). - Hooks and documentation updated to clarify there is no network upload path and that replayed excerpts are always untrusted. - Lifecycle hooks now check for CLAUDE_PLUGIN_ROOT and exit early if set. - Removed the sample skill-card.md file.

- Removed the redundant skill-card.md file. - No changes to functionality or documentation; disk usage slightly reduced.

- Improved script and docs placement: added standalone examples.md, reference.md, and moved templates to a top-level /templates folder. - Added new analytics-specific templates: analytics_findings.md and analytics_task_plan.md. - Updated PowerShell and shell command logic to handle cross-platform stop hooks more robustly. - Clarified session restoration instructions in documentation. - Removed legacy/duplicative files from the references and root directory for cleaner structure.

- Expanded Windows and Linux compatibility for the "Stop" hook with improved script selection logic. - Environment variable now auto-detects the correct skill directory for session recovery, supporting both plugin and default install paths. - Documentation streamlined and clarified, with updated setup and context restoration instructions. - Added scripts/__pycache__/session-catchup.cpython-314.pyc for internal use; removed the obsolete skill-card.md file.

Major update in v3.0.0: hooks migrated to external reusable shell scripts, improving modularity and maintainability. - All bash logic for hooks moved from SKILL.md into standalone scripts in the scripts directory. - Added new scripts: inject-plan.sh, gate-stop.sh, ledger-append.sh/ps1, ledger-summary.sh/ps1, phase-status.sh/ps1. - PostToolUse and Stop steps now call reusable scripts instead of embedding commands. - Obsolete/inlined content (skill-card.md) removed; new task_plan_autonomous.md template added. - Greatly clarifies the distinction between skill assets and user project files.

No user-facing changes in this version. - No changes to files or behavior detected.

No user-facing changes in this version. - No file changes detected. - Functionality and interface remain the same as the previous release.

No user-facing changes in this release. - No file changes were detected between versions. - Functionality and behavior remain unchanged.

- Improved plan file resolution: now supports resolving and loading plans from .planning subdirectories with more robust logic. - Enhanced attestation (tamper detection): SHA256 caching is used for faster plan integrity checks. - Progress and attestation checks now work with named/scoped plans as well as root plans. - PreToolUse and UserPromptSubmit hooks updated to always find the correct plan, even if not in root. - No changes to external interface or required user actions.

- Minor internal update: output markers in hooks changed from "---BEGIN PLAN DATA---" to "===BEGIN PLAN DATA===" for consistency. - No changes to user workflow or visible features. - Version bump to 2.38.1, no file changes detected.

**Introduces file integrity attestation and tamper-detection for planning files.** - Added plan attestation and SHA-256 verification to detect tampering in task_plan.md before it is used, improving security against indirect prompt injection. - New hooks now block processing of a plan if the hash does not match the attestation file, and display clear messages guiding re-attestation or restoration. - Added support scripts for managing attestation and plan directories, including PowerShell and shell variants. - Expanded file automation with helper scripts: init-session, check-complete, set-active-plan, and resolve-plan-dir. - Updated version to 2.38.0 and extended hooks to include PreCompact for context compaction events.

**No significant user-facing changes.** - Updated the `Stop` hook to use a less restrictive PowerShell execution policy (`RemoteSigned` instead of `Bypass`). - Bumped version to 2.36.1. - No behavior or documentation changes affecting general use.

**Changelog for planning-with-files v2.35.0** - Improved platform compatibility for the Stop hook by updating script location logic and fallback. - Minor fixes to allowed-tools formatting. - No code changes; all planning methodology and core behavior remain unchanged.

planning-with-files v2.34.0 Codex users — this one's for you. The lifecycle hooks that @Leon-Algo built in v2.31.0 were accidentally wiped during the v2.32.0 release cycle. All five are back: SessionStart runs session catchup and injects your active plan; UserPromptSubmit re-injects on every message so the agent never forgets where it left off; PreToolUse re-reads task_plan.md before touching your codebase; PostToolUse reminds the agent to log what it just did; Stop blocks when phases are still open and re-prompts to keep going. Codex now has full hook parity with Claude Code, Cursor, and Copilot. session-catchup.py is also updated with the complete Codex rewrite — structured patch_apply_end event detection, session size filtering, and current-thread preference via CODEX_THREAD_ID. Everything else in this release is cleanup. Two shell scripts that should have been executable weren't — fixed, with a regression test so it never happens again. Four contributors who shipped real work never got credited — they're in CONTRIBUTORS.md now. A duplicate metadata key was lurking in several SKILL.md variants — gone. The Tessl CI workflow is in, which means every future SKILL.md PR gets automatically scored and reviewed. 154 of you are using this. Thank you for staying with it.

planning-with-files 2.22.1 - Added Windows PowerShell scripts: scripts/check-complete.ps1 and scripts/init-session.ps1 for Windows compatibility. - Improved cross-platform support for session initialization and completion checks. - No changes to planning workflow or core features.

FOLDER TO UPLOAD: .openclaw/skills/planning-with-files/ LICENSE: Check the box "I have the rights to this skill and agree to publish it under MIT-0" === CHANGELOG TEXT (copy this into the CHANGELOG field) === v2.22.0 - Major update from v1.2.0 Security: - Added Security Boundary section to prevent prompt injection via task_plan.md - Removed WebFetch/WebSearch from allowed-tools (these don't belong in a planning skill) - External content must go to findings.md only, never task_plan.md Features: - Formal benchmark results: 96.7% pass rate (29/30 tests) vs 6.7% without skill - A/B comparison wins: 3/3 (100%), avg score 10.0/10 vs 6.8/10 - Session recovery via session-catchup.py for context continuity - Cross-platform support (macOS, Linux, Windows) Fixes: - Fixed Windows PowerShell 5.1 compatibility issues - Fixed Stop hook multiline YAML parsing on Git Bash - Fixed UTF-8 encoding issues in hook scripts - Corrected broken documentation links This skill implements Manus-style file-based planning using task_plan.md, findings.md, and progress.md for persistent working memory.

**v1.2.0 — Cursor hooks, script fixes, multi-IDE improvements** - Added Cursor hooks support (preToolUse, postToolUse, stop with auto-continuation) - Fixed check-complete scripts to support both **Status:** and [bracket] status formats - Fixed stop hook to always exit 0 (no more false errors in Claude Code) - Fixed plugin install command to use correct slash commands - Added session-catchup.py for context recovery across sessions - Updated docs for Cursor, Kilocode, and quickstart - Renamed Moltbot references to OpenClaw - Added Plan Cascade to community forks - Now supports 14+ IDEs including Pi Agent, OpenClaw, and Cursor with native hooks

- Updated skill.