to-tasks
Split each ticket into implementation tasks that write to at most two projects, nested under the ticket.
Skill by GUOJIE-HONG
Chalk Labs
Need help deploying to-tasks?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/guojie-hong-to-tasks
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: design, writing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/to-tasks
What's inside
- Skill name
to-tasks- Description
- Split each ticket into implementation tasks that write to at most two projects, nested under the ticket. Use after to-tickets once design.md records the chosen direction, before any code is written.
- License
- MIT
- Metadata
- internal: true
- Instructions
- 135 lines in SKILL.md · view on GitHub
Files (2)
- SKILL.md
- openai.yaml · agents/
SKILL.md (first lines)
# To Tasks A ticket from `to-tickets` is a tracer bullet: one end-to-end behaviour, deliberately layer-agnostic. Implementing it usually means creating or editing files across several projects at once, and that is where a small model in a long session drifts. This skill turns each ticket into **tasks**: implementation slices whose **footprint** (the projects they write to) is at most two projects, ordered along the dependency graph, each a short step the ticket's worker finishes, validates, and commits before starting the next. Tasks nest under their ticket: ```text ticket 01 ├─ task 01 ├─ task 02 └─ task 03 ``` This skill writes task files only. It ends when the tree is published. ## 1. Establish the input The input is a feature directory holding the tickets and `design.md`. Accept it as a path or infer it from the conversation; when the conversation already holds the tickets and the design, work from that. Read, in this order: 1. The issue tracker convention (`docs/agents/issue-tracker.md` or whatever `/setup-matt-pocock-skills` configured). It decides where tasks land in §6. 2. `design.md` in the feature directory. It carries the chosen direction and the conventions to follow, each with an evidence path. Those paths are the raw material for task footprints. 3. Every ticket file. Note each ticket's number, blocking edges, and acceptance criteria. 4. `spec.md`, when present, for the behaviour the acceptance criteria refer to. `design.md` missing means the direction is undecided. Say so and stop; splitting an undecided design produces tasks that will be thrown away. Do not ask the user for facts discoverable from the repository or these files. ## 2. Map the project graph Read the repository's project definitions once and record two things: the list of projects, and the dependency direction between them. Where to look, by ecosystem: - .NET: every `*.csproj` and its `ProjectReference` entries; the `.sln` for the full set. - JavaScript / TypeScript monorepos: the workspace globs in `package.json` or `pnpm-workspace.yaml`, then each package's dependencies on sibling packages. - Go: `go.mod` modules and `go.work`. - Anything else: the build system's unit of compilation that has its own dependency declaration. Projects `design.md` says to create do not exist on disk yet. Add each to the list with the dependencies `design.md` gives it, citing that line; a new project whose dependencies `design.md` leaves open is a gap to raise in §5, not one to guess. A **project** is a unit that declares its own dependencies. Test projects are projects. When the repository is a single project, the boundary is the top-level directory that a module system or the build treats as a unit; say which rule you applied. Record the graph in the report as a short chain or list, inner-most (no dependencies) first. Every later step orders tasks along it. ## 3. Compute each ticket's footprint For each ticket, list the projects that must be created in or edited to deliver its acceptance criteria under the direction in `design.md`. For every project in the footprint, cite the convention in `design.md` (with its evidence path) or the ticket line that puts it there. A project with no citation is not in the footprint. Include the test project that proves each production change. The test project counts toward the footprint like any other. Sort the footprint inner-most first, following the graph from §2. ## 4. Split into tasks Walk each ticket's sorted footprint and cut it into tasks: - **Footprint limit**: each task writes to at most two projects. A production project and the test project that proves it is the usual pair. - **Dependency order**: an earlier task never depends on a later one. Give each task its **blocking edges** among tasks of the same ticket. Cross-ticket order is already carried by the ticket's own blocking edges; tasks reference only siblings. - **One validation**: each task names the single narrowest command that proves it: a build of the affected project, one test file or filter, one migration command, one HTTP call. Take the commands from `design.md`, the repository's instruction files, or the test project's own documentation; a command you cannot ground stays out. - **Checkpoint size**: each task must be finishable, validated, and committed as one step; its commit is where a worker that ran out of context resumes. When a two-project slice is still too large for one step, split it by behaviour inside the same footprint. - **Concrete**: tasks name files, symbols, and behaviour. This is the reverse of the ticket rule; tickets avoid paths because they outlive the code, tasks are executed next and their paths come straight from `design.md`. A ticket whose whole footprint already fits in two projects becomes exactly one task, carrying the same concrete detail. Downstream then sees one shape for every piece of work. The last task of a ticket is the one that makes the ticket's acceptance criteria observable; say so in that task so the executor knows to run the ticket's criteria after its own validation. ## 5. Quiz the user Present the tree as a nested list. For each task show: - **Title** - **Writes to**: the one or two projects - **Blocked by**: sibling task numbers, or none - **Validation**: the one command Ask: - Does any task's footprint look wrong: a project missing, or one that should not be there? - Are the blocking edges the real dependencies? - Should any task be merged or split further? Iterate until the user approves the tree. Nothing is written before that approval. ## 6. Publish the tasks Publish in the shape the configured tracker uses: - **Local files** → one file per task under `<issues dir>/<NN>-<ticket-slug>/<MM>-<task-slug>.md`, numbered from `01` in dependency order. The directory name matches the ticket file's basename, so the ticket and its tasks sit side by side. - **A real tracker** → one sub-issue per task under the ticket's issue, in dependency order, using the platform's native blocking relationship; otherwise set "Blocked by" to the sibling issues. Apply `ready-for-agent`. Leave the ticket files untouched. <task-template> # <NN>.<MM>: <Task title> **Parent ticket:** <NN>-<ticket-slug> **Blocked by:** <sibling task numbers>, or "None (can start immediately)" **Status:** ready-for-agent ## Boundary Every edit lands in these projects: `<Project A>`, `<Project B>`. Read any project; write only inside this ticket's footprint, the projects its tasks name. A change an earlier sibling missed in its own project is fixed in place and noted under that sibling's Comments. When the work turns out to need a change in a project no task of this ticket names, stop, record which project and why under Comments, set Status to needs-triage, and leave the change unmade. ## What to change Files to create or edit, the symbols involved, and the behaviour each must have. Cite the `design.md` convention (with its evidence path) behind each non-obvious choice. ## Validation One command. ## Done when - [ ] Validation passes - [ ] <observable outcome specific to this task>
From GUOJIE-HONG/skills, MIT.
More from GUOJIE-HONG/skills
- 0.589score
dont-know-how
GUOJIE-HONG
Find direction for a task you don't know how to start.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
design-code-implement
GUOJIE-HONG
Turn a settled requirement into at least three grounded implementation directions, then record the chosen one.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
open-course
GUOJIE-HONG
Turn a topic into a Course project, a Syllabus plus a Handout and Slides per Lesson, where every factual claim cites a Source read for this Course and a point…
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
show-grill-clearly
GUOJIE-HONG
Turn the unresolved decisions of a grilling session into a temporary browser questionnaire, so questions that are hard to follow in chat become clear, while…
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
ask-jev
GUOJIE-HONG
Let TypeSafe's Jev settle the Open questions your sources leave, and collect the rest into a questionnaire for the person who decides.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
clean-comments
GUOJIE-HONG
Rewrite the existing code comments in a repo or path so each says only what the code is for, and list what needs the user's judgement (comments that disagreed…
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
torture-gently
GUOJIE-HONG
Grill the user about a plan, decision, or idea along grounded, decision-relevant branches.
SkillAFreshness: A★ 1Claude Code +9 - 0.586score
refactor
GUOJIE-HONG
Restructure code under a plan, test-first when behavior changes, against a green baseline when it must stay the same.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
seen-it-before
GUOJIE-HONG
Name the classic problems a hard problem is a variant of, with their known solutions and where the problem departs from them.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
implement-task
GUOJIE-HONG
Execute to-tasks tickets by dependency order, one fresh sub-agent per ticket in its own worktree, independent tickets in parallel, each ticket merged back and…
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
implement-all
GUOJIE-HONG
Implement a whole spec on one branch, its tickets in parallel sub-agents following design.md, ending in a merge request on the repository's own host.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
ptns
GUOJIE-HONG
Prompt to new session — hand the current progress to a fresh session as one short prompt, delivered by session-to-session message on Claude Code, or as a…
SkillAFreshness: A★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 1
- Open issues
- 1
- Last commit (this folder)
- Oct 7, 2026 (24 days ago)
- Last release
- Oct 23, 2026 (8 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/GUOJIE-HONG/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from GUOJIE-HONG
- 0.589score
dont-know-how
GUOJIE-HONG
Find direction for a task you don't know how to start.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
show-grill-clearly
GUOJIE-HONG
Turn the unresolved decisions of a grilling session into a temporary browser questionnaire, so questions that are hard to follow in chat become clear, while…
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
design-code-implement
GUOJIE-HONG
Turn a settled requirement into at least three grounded implementation directions, then record the chosen one.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
ask-jev
GUOJIE-HONG
Let TypeSafe's Jev settle the Open questions your sources leave, and collect the rest into a questionnaire for the person who decides.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
clean-comments
GUOJIE-HONG
Rewrite the existing code comments in a repo or path so each says only what the code is for, and list what needs the user's judgement (comments that disagreed…
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
open-course
GUOJIE-HONG
Turn a topic into a Course project, a Syllabus plus a Handout and Slides per Lesson, where every factual claim cites a Source read for this Course and a point…
SkillAFreshness: A★ 1Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get to-tasks running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
to-tasks skill. Agent Mart. https://agentmart.dev/skills/guojie-hong-to-tasks (last updated 2026-10-24)