- Home
- Skills
- Productivity
- Planning and Task Breakdown Playbook
More screenshots
Works with the AI Agents you already use
Planning and Task Breakdown Playbook
Turn a spec into an ordered task list that survives contact with reality: outcome-titled tasks with checkable acceptance criteria, named verification, explicit dependencies, and checkpoints.
$5
Planning and Task Breakdown Playbook
Example session with this skill installed
Break this into tasks: add CSV export to our reporting page. Users pick a date range and get a CSV download. Exports over 10,000 rows run in the background and email the user when the export is ready.
- Read your context and instructions
- Compiled the planning and task
Task 1: Report query returns a caller-chosen date range
The existing report query gains start/end parameters used by both the page and the export. Criteria: page results match the selected range exactly; an empty range returns zero rows, not an error; existing saved reports still load. Verification: report test suite plus a manual query against the seeded dev data. Depends on: none. Scope: query layer only — no UI, no export yet.
Task 2: CSV serializer streams rows in report column order
Criteria: output columns match the on-screen report order and headers; quoting handles commas and newlines in fields; a 50k-row fixture serializes without loading it fully into memory. Verification: serializer unit tests including the quoting fixture. Depends on: none. Parallel-safe with Task 1.
Task 3: User downloads a CSV for ranges up to 10,000 rows
Criteria: Export button downloads a file named with the report and range; a 10,000-row export completes as a direct download; row count in the file matches the on-screen total. Verification: end-to-end test on the reporting page with seeded data. Depends on: 1, 2.
CHECKPOINT after Task 3 — suite green; the small-export path works end-to-end in the dev environment; the 10,000-row threshold is confirmed as the product decision, not a guess.
Task 4: Oversized exports are queued as background jobs
Criteria: a 10,001-row export creates a job instead of downloading; the page shows queued/running state; duplicate clicks don't create duplicate jobs. Verification: job tests plus manual run against the 50k-row fixture. Depends on: 3.
Task 5: User is emailed a download link when a background export finishes
Criteria: email arrives with a working link; the link expires after 48 hours; a failed job emails nothing and surfaces the failure on the page. Verification: mailer test with a captured inbox, link-expiry test. Depends on: 4.
NOT in this list: scheduled/recurring exports, Excel format, admin visibility into other users' exports — noted as follow-ups, not oversights.
Connects securely to your tools. The creator never sees your data.
About this skill
Bad task breakdowns fail in predictable ways — tasks that describe activities instead of outcomes, dependencies nobody wrote down surfacing mid-work, tasks sized for optimism, and ten "completed" tasks that were never once verified. This skill replaces all of that with a mechanical discipline. Every task is an atom: an outcome title ("User receives email on report completion," not "email stuff"), a one-or-two-sentence description, 2–4 acceptance criteria a specific person could check yes or no, the exact verification command or manual check, explicit dependencies, and a scope note naming what's not in the task. The method reads the spec and the actual code first, draws the dependency graph (the graph dictates the order, not preferences), and slices vertically so each task delivers something observable end-to-end instead of deferring all surprises to a horizontal layer cake. Sizing follows the one-session rule — completable in one focused sitting, a handful of files — with a decision table for splitting tasks that fail the heuristics: titles with "and," criteria lists that won't stop growing, one unknown buried in routine work (split out a spike whose deliverable is a decision). Ordering is dependencies first, then risk, then value as tiebreaker, with parallel-safe tasks marked and shared interfaces defined as contract tasks before parallel work starts. Checkpoints sit where reality gets checked, every few tasks, with pass conditions. Each drafted task is scored on a five-point quality rubric — outcome title, criteria, verification, dependencies, size — and the list doesn't ship until every task scores full marks. The plan lives in a file, not in chat, because sessions end and context compacts. Edge cases are covered: specs changing mid-breakdown, not-yet-possible work, inherited task lists, and unowned external dependencies.
- It decomposes; it does not decide. If the spec's outcomes aren't agreed yet, the skill stops and asks rather than producing confident-looking tasks on sand.
- Verification steps use the test and build commands you supply — if the project has no runnable checks, criteria fall back to named manual checks, stated as such.
- Sizing heuristics assume focused solo or small-team work; they calibrate sequencing and verifiability, not calendar commitments.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected