Daily Work Planner

Daily Work Planner

This is the Codex/OpenAI skill entrypoint. For agents that read generic instruction files instead of SKILL.md, use AGENTS.md in this same folder.

Core Rule

Optimize for execution, not planning. Convert uncertainty into a usable first plan quickly, then use checkpoints to correct the plan during work.

Before planning, calculate the planning budget:

planning_budget = min(available_work_time * 5%, absolute_cap)

Use these caps:

• <= 60 minutes total work: max 3 minutes planning
• 61-240 minutes total work: max 8 minutes planning
• 241-480 minutes total work: max 15 minutes planning
• Multi-day or > 480 minutes: max 20 minutes for the first planning pass

If the planning budget is small, produce a minimal plan instead of asking many questions.

Workflow

1. Identify the task goal, current-window task text if present, available time, hard deadline, expected output, local repo, and provided files.
2. Estimate required minutes separately from available minutes. If a hard deadline exists, calculate the actual start-to-deadline window; never substitute the required-time estimate for available time.
3. Score feasibility by comparing the actual available window with the required-time estimate. Use scripts/feasibility_score.py to choose complete, deliverable, minimum, or redefine scope.
4. Calculate planning budget and choose output depth: quick, standard, or full.
5. Inspect only enough source material to determine scope and priority. Use scripts/extract_file_context.py for local PDFs, DOC/DOCX, PPTX, XLSX, Markdown, text, and code when deterministic file context is useful.
6. Choose or classify the work mode. Use scripts/classify_work_mode.py when the mode is unclear.
7. Classify files as primary, reference, optional, or ignore-for-now. Use scripts/rank_files.py when multiple files are provided.
8. Set hard deadline, soft deadline, buffers, and checkpoint times.
9. Split the work into milestones with acceptance criteria.
10. Define the minimum deliverable version.
11. Add a delay-response plan for checkpoint slippage.
12. During execution, use scripts/checkpoint_session.py to record progress and detect whether the plan should be compressed.
13. If the user resumes later, use scripts/resume_session.py to create a concise resume card before replanning.
14. At the end, use scripts/handoff_session.py to create a handoff and optionally write local memory.
15. Validate the plan with scripts/validate_plan.py when a durable plan is being saved or shared.
16. For complete work-session packages, persist session.json and update it through scripts/session_state.py.
17. After completion or review, record a sanitized local memory entry with scripts/task_memory.py when the user wants task history, estimate calibration, or habit memory. Support list, forget, purge, retention, and export operations.
18. Output the plan in a compact form the user can execute immediately.

Source Handling

Respect the planning budget when reading files.

• For large PDFs or documents, inspect titles, headings, tables of contents, abstracts, filenames, metadata, and representative pages first.
• Avoid deep summarization unless the user's task is explicitly reading or analysis.
• Prefer local file inspection. Do not upload sensitive files.
• If files are too many, prioritize by filename, modified time, user-stated relevance, and relationship to the final deliverable.
• Do not claim to read private OS windows directly. In Codex, use the current conversation/context that is already available; in scripts, accept copied window text through --window-note.

Output Depth

Use Quick Plan for short or unclear tasks:

• Goal
• Planning budget
• First action
• 2-4 steps
• Soft and hard deadline
• Minimum deliverable

Use Standard Plan for most same-day work:

• Task summary
• Planning budget
• File priority
• Soft and hard deadline
• Milestone table
• Buffer
• Minimum deliverable
• Delay response

Use Full Plan only for complex full-day or multi-day work:

• Scope decision
• File priority table
• Milestones and acceptance criteria
• Checkpoints
• Buffer and fallback paths
• Todo list
• Review log template
• Optional ICS event instructions
• Consolidated TXT/DOCX report for human reading

References

Load these only when needed:

• references/planning_rules.md: budget, soft deadline, buffer, and output-depth rules.
• references/work_modes.md: work-mode-specific planning rules.
• references/reschedule_rules.md: checkpoint delay handling and minimum deliverable rules.
• references/privacy_rules.md: local-first and sensitive-content handling.
• references/memory_rules.md: local task memory, habit learning, and what not to store.

Scripts

Use scripts when deterministic output is helpful:

• scripts/plan_day.py: create a Markdown work-session plan from start time, total minutes, mode, goal, and optional deadline.
• scripts/extract_file_context.py: extract lightweight local context from PDF, DOC/DOCX, PPTX, XLSX, Markdown, text, and code files for priority decisions.
• scripts/inspect_tasks.py: inspect current-window notes and local repositories for open tasks, then estimate speed-adjusted required minutes.
• scripts/feasibility_score.py: score whether the work fits the time box and choose complete, deliverable, minimum, or redefine scope.
• scripts/classify_work_mode.py: infer the most likely work mode from the goal and file names.
• scripts/rank_files.py: rank files as primary, reference, optional, or ignore-for-now.
• scripts/validate_plan.py: check that a plan includes deadlines, milestones, acceptance criteria, buffer, fallback, and budget compliance.
• scripts/start_session.py: generate a full work-session package with consolidated work_session.txt, work_session.docx, session.ics, and session.json; use --split-files to also write separate helper files.
• scripts/make_todo.py: convert a milestone table into a todo.txt style checklist.
• scripts/make_ics.py: create timezone-safe calendar .ics events for the session and its milestones, optionally with a display reminder.
• scripts/report_writer.py: write consolidated planning reports as TXT and DOCX without third-party dependencies.
• scripts/update_review_log.py: append planned-vs-actual records to a Markdown review log.
• scripts/reschedule_session.py: rebuild the remaining work plan after checkpoint slippage.
• scripts/checkpoint_session.py: record done/remaining work during execution and classify delay severity.
• scripts/resume_session.py: generate a resume card from session.json and local memory.
• scripts/handoff_session.py: create an end-of-session handoff and optionally update local memory.
• scripts/estimate_profile.py: summarize review logs to calibrate future time estimates.
• scripts/session_state.py: create, inspect, transition, and review durable session.json state.
• scripts/task_memory.py: append sanitized local task memory, generate a private summary, and list, forget, purge, retain, or export entries.

Output Requirements

Every plan must answer:

1. What should the user do first?
2. What counts as done for each milestone?
3. What should be protected if time runs out?

Keep the final plan compact. Never let planning consume more value than it saves.