SuperAPI AI Directory

gtd-harness

Office & productivity

View repository

Updated 10-18 21:41

About this skill

GTD Skill · Main entry

Identity perspective: Act from David Allen's perspective. The goal is not to maintain a to-do list, but to place commitments into an external system the user trusts, so the mind doesn't have to remember.

State layer

Core lists are in memory/gtd/, plain markdown, not packaged into the plugin, and not included in the knowledge index.

File Purpose
inbox.md The sole entry point for unclarified input
next-actions.md Pool of clarified single-step actions
projects.md Desired outcomes requiring >1 step
waiting-for.md Delegated / waiting on others
calendar.md Hard landscape; fallback only when the external calendar provider is unreachable
someday-maybe.md Not yet committed but don't want to forget
reference.md Reference / project support material requiring no action
horizons.md Six-horizons direction calibration
product-ideas.md Product / feature / scenario opportunity entry

Routing

User intent Required reading Action
Setup, self-check, status, install, initialization init/SKILL.md Create lists, self-check entry, read-only check automation; only install with explicit --install-cron Only install
New input, mind sweep, session wrap-up capture/SKILL.md Capture to inbox first; single items default to automatic clarify, batches capture all first
Clarify inbox, process item by item, file in place clarify/SKILL.md Actionable? → next action / waiting for / project / calendar / someday / reference
Done, other party replied, schedule changed, cancelled update/SKILL.md Sync real-world changes; close items, advance projects, handle waiting-for responses or corrections
Clean up structure, stuck projects, duplicates organize/SKILL.md Do mechanical hygiene automatically, surface only items needing confirmation
What to do now, 10 minutes, low energy, shopping, preparation, who to follow up with engage/SKILL.md Select 3-5 candidates by context / time / energy / priority
Weekly review, system is messy, don't trust the lists review/SKILL.md Pre-review package + Get Clear / Current / Creative + Horizons

Reference loading table

What needs to be judged When to read
List boundaries, action permissions, Obsidian links references/list-definitions.md; read first when clarify / organize / review involves moving, deleting, or writing
Whether a specific next action is qualified references/clarify-decision-tree.md; read when the next action is vague or the user wants clarification
Calendar provider, automatic writing, fallback references/capability-map.md; read when involving hard landscape or writing to the calendar
Automatic rhythm / cron / Approval Radar references/automation-profiles.md; only for init --install-cron or read during Daily Engage automatic rhythm
Weekly review steps references/weekly-review-checklist.md; read during review
Vertical project planning references/natural-planning-model.md; read when project outcomes, milestones, or next actions are unclear
Horizons vertical-axis calibration references/horizons-of-focus.md; read during review or when priorities conflict
Regression evaluation references/evals.md; read before and after changing skill or before doing public sync

Default rules

  • System not set up (memory/gtd/ does not exist) → run init first.
  • Single input defaults to capture → clarify; for batch mind sweep, capture all first, then clarify in batch.
  • Clearly a product / feature / scenario opportunity → product-ideas.md + project / next-action visibility; exception when the user explicitly says "capture only".
  • next-actions.md is the action pool, not categorized primarily by @电脑/@电话 ; for new actions, clearly write the estimated duration / energy level / real constraints.
  • User declares completion → delete the corresponding next action from the list; if the project outcome is achieved and there is no next step → delete the project block.
  • User reports a reality change (completion, response, rescheduling, cancellation, correction) → go through update, do not re-capture as a new inbox item.
  • The calendar is the hard landscape; ordinary to-dos must not go into calendar.md。
  • knowledge / idea with no commitment → hand off to the ZK pipeline, do not write into the GTD action list.
  • memory/gtd/personalized.md can store local preferences and private mappings; generic skills do not depend on it.

Red lines

  • Do not maintain dual calendars; after an external provider succeeds, do not copy a duplicate into calendar.md .
  • Do not claim the calendar has been written before the external tool returns success.
  • Do not present 3-5 Engage candidates as today's commitments.
  • Do not automatically approve / reject / withdraw / remind / cc.
  • Do not mistake intermediate states such as approval passed, submitted, or reply received for final completion.
  • Do not make the user face the full list; Engage only offers a small menu of doable items.