task-workflow
Shared task ledger, gate, artifact, and Git-safety contract for the canonical task-start workflow.
Skill by blakesims
Chalk Labs
Need help deploying task-workflow?
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/blakesims-task-workflow
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: documents, productivity
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/task-workflow
What's inside
- Skill name
task-workflow- Description
- Shared task ledger, gate, artifact, and Git-safety contract for the canonical task-start workflow.
- License
- MIT
- Instructions
- 88 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Task Workflow `tasks/` is the durable record of human intention, agent planning, execution, and review. The entry point is `/task-workflow:task-start`; this file is the shared contract it and every child agent operate under. ## Task creation 1. Read `CLAUDE.md`, this file, and `tasks/global-task-manager.md`. 2. Get `Next ID` from the GTM, create `tasks/planning/TXXX-task-slug/`, copy `tasks/main-template.md` to `main.md`, add the task to the GTM, and increment `Next ID`. 3. Draft the Intent Contract and `DONE_WHEN` from the human's request; set status `PLANNING`. 4. Offer `/task-workflow:intent-harden` for non-trivial work and record the outcome on the `Intent hardening:` line in `main.md`. If it runs, the hardened brief replaces the draft contract. 5. Build the Handoff Packet (defined in `task-start`). Every child prompt receives the same `DONE_WHEN` verbatim. 6. Route the task: `quickfix` (small, one plausible site, obvious validation, no product ambiguity — planning skipped, code review kept) or `planned`. Record the lane in `main.md`. ## Status machine | Status | Meaning | Directory | |---|---|---| | `PLANNING` | Intent/plan being prepared | `tasks/planning/` | | `PLAN_REVIEW` | Plan reviewer checking readiness | `tasks/planning/` | | `READY` | Plan passed; task can execute | `tasks/active/` | | `EXECUTING_PHASE_N` | Executor implementing phase N | `tasks/active/` | | `CODE_REVIEW` | Code reviewer checking phase N | `tasks/active/` | | `BLOCKED` | Human input or prerequisite required | `tasks/paused/` | | `COMPLETE` | All phases implemented and reviewed | `tasks/completed/` | A quickfix enters `tasks/active/` at routing and runs as a single phase. The parent moves folders with a normal filesystem move, keeps the `## Meta` status and GTM row in agreement, and stages the old/new paths at the next approved commit boundary. Tasks created before this workflow are historical evidence: reassess their intent against current repository direction before resuming; never execute stale plans automatically. ## Approved plan and report pointers Keep the entire approved Task, Intent Contract and phased `## Plan` in `main.md` as the historical what/how/where agreement. Do not amend it, tick its checkboxes or append execution/review material. Agents write separate reports first, then replace compact status/outcome/path entries outside the specification; this is not a cumulative line quota. - Executor: `execution-phase-N.md` → current phase under `## Execution Log`. - Reviewer: `code-review-phase-N.md` → gate/date/path under `## Code Review Log`. Phase reports identify phase, execution attempt and baseline. `final-review.md` identifies cumulative baseline, current working tree, and review attempt, linked by a compact final-review entry. - Plan reviewer: preserve numbered attempts in `plan-review.md`, recording `Review attempt` and `Reviewed specification SHA-256` for the exact Task/Intent/entire Plan byte range defined in task-start. Before `READY`, the parent compares latest report/pointer/dispatched attempt and freshly computed current digest; missing/stale evidence blocks, including after retries/replans. Metadata stays outside the specification. - On `REVISE`, pass the exact current full review path; the executor must read its numbered blockers before repair. Parent checks report existence and current attempt before routing gates. - Parent: status/GTM and compact completion pointer; full completion evidence in `completion.md`. - If evidence invalidates scope, record it separately and stop `BLOCKED` for explicit replanning authorization and renewed review/acceptance. Preserve the earlier agreement in Git/history. - Do not automatically rewrite historical tasks or upgrade copied templates without consent. Agree on the authoritative plan revision before resuming; preserve old records and use separate reports for new work. Before routing gates, compare the full approved specification against its accepted Git revision and original path; only declared metadata/pointers may change. These are prompt-level safeguards, not a filesystem sandbox. Tests cover instructions/artifact fixtures, not guaranteed model compliance. ## Ownership and gates | Artifact | Owner | |---|---| | Intent Contract and Handoff Packet | Human + parent orchestrator | | `## Plan` | `task-workflow:planner` | | `## Plan Review` / `plan-review.md` | `task-workflow:plan-reviewer` | | Source edits, `execution-phase-N.md`, and compact `## Execution Log` pointer | `task-workflow:executor` | | `## Code Review Log` / `code-review-phase-N.md` | `task-workflow:code-reviewer` | | Git commits after `PASS`, GTM, transitions | Parent orchestrator | | Gate | Outcome | Action | |---|---|---| | Plan review | `READY` | Promote to `tasks/active/` and execute | | Plan review | `NEEDS_WORK` | Planner revises, re-review | | Plan review | `NOT_READY` | Block for a real human decision/prerequisite | | Code review | `PASS` | Parent commits the reviewed phase and continues | | Code review | `REVISE` | Executor repairs numbered findings, re-review | | Code review | `FAIL` | Block or re-plan | Maximum three `NEEDS_WORK` or `REVISE` cycles before blocking. If a reviewer returns a verdict without persisting its declared artifacts, the parent writes the exact output to the separate report and updates only its outcome/path pointer before routing the gate. ## Review scope and safe validation - Block on concrete defects, unmet criteria or material evidence gaps. Tooling/editorial suggestions do not force revision unless they affect executable requirements, required evidence or safety decisions. Do not waive approved criteria. - Re-review prior blockers, repair delta and affected regressions/dependencies. Keep the full contract, cumulative integration obligations and freshness checks; reference still-valid evidence and expand review when new evidence undermines it. - Choose checks by risk, not quotas for findings, reruns or negative controls. Inspect test configuration before running. Source-mutating controls must run only in a disposable isolated copy of the exact candidate, including uncommitted and untracked candidate files, with no shared writable source or live-service side effects. Never run them in the authoring tree; report unavailable safe verification as a gap. Required completion checks remain required, safely isolated when necessary. - Preserve reports/attempts. Record blocker dispositions, changed evidence and prior report/attempt pointers, not repeated history. Handoffs carry gate, decisions, current attempt/baseline and exact paths; write full evidence once. ## Git and substrate discipline - Choose current branch, feature branch, or worktree from repository instructions, Git state, concurrency, and delivery expectations; record the strategy, rationale, and task baseline SHA in `main.md`. - Preserve unrelated human changes: never stash, reset, delete, or commit them; prefer safe isolation when allowed. - Require a clean workspace before phase preparation. The executor may then see only the declared parent status/pointer edits; on repair it resumes the expected reviewed diff under the original phase baseline. Unrelated changes still block or require isolation. It leaves its phase uncommitted for review. - The code reviewer inspects the actual working-tree diff from the recorded phase baseline, runs checks, and persists its verdict; the plan reviewer does the same for plans. - The parent commits only after `PASS`, using `git add -- <explicit-reviewed-paths>` and a task/phase-specific message. Never use `git add .`, `git add -A`, or stage a path the reviewer did not inspect; verify `git diff --cached --name-only` before every commit. - Push, PR, merge, deployment, force-push, and branch deletion require explicit authorization or an already-authorized repository workflow. ## Blocked recovery Record the exact blocker and attempted work in the current phase report (or `blocked.md` before execution), link it from a compact blocker/status entry in `main.md`, set `BLOCKED`, move the task to `tasks/paused/`, and update the GTM. When the human answers, restore the appropriate prior stage — an answer does not imply "resume coding"; return to planning if intent or scope changed. ## Completion Run the repository's full tests/lint/build checks under the safe-validation rules above, compare the cumulative result with `DONE_WHEN`, write full evidence to `completion.md` and update only the outcome/date/path under `## Completion`, move the task to `tasks/completed/`, update the GTM, and commit the ledger update. Follow the recorded delivery strategy; do not push, merge, open a PR, or deploy unless authorized.
From blakesims/task-workflow-plugin, MIT.
More from blakesims/task-workflow-plugin
- 0.524score
cca-task-shape
blakesims
Shape a real task before execution.
SkillBFreshness: BClaude Code +9 - 0.524score
doc-new-note
blakesims
Use when asked to create a "new note", "daily summary", "worklog entry", "weekly summary", or similar, in a project that has a worklog system…
SkillBFreshness: BClaude Code +9 - 0.524score
intent-harden
blakesims
Stress-test and tighten an Intent Contract before planning.
SkillBFreshness: BClaude Code +9 - 0.521score
task-start
blakesims
Run the task-workflow development cycle — Intent Contract, routing, reviewed phased execution — whenever the user asks to use the task workflow or task system…
SkillBFreshness: BClaude Code +9 - 0.521score
doc-init
blakesims
Set up a lightweight project documentation system — worklog (transient notes) plus refs (curated docs).
SkillBFreshness: BClaude Code +9 - 0.521score
doc-sweep
blakesims
End-of-day synthesis — reads today's worklog notes, resolves cross-note tensions, and fills today's daily summary.
SkillBFreshness: BClaude Code +9 - 0.521score
investigation-review
blakesims
Independently review a debugging conclusion against risk-critical evidence; return PASS or REVISE with concrete numbered blockers, separate from suggestions.
SkillBFreshness: BClaude Code +9 - 0.521score
linear-independence
blakesims
Use whenever you are presenting the user with questions or options to reduce cognitive load and complexity.
SkillBFreshness: BClaude Code +9 - 0.521score
scientific-method
blakesims
Use hypothesis-driven investigation when uncertainty could change a conclusion or safety decision.
SkillBFreshness: BClaude Code +9 - 0.520score
task-management-init
blakesims
Set up the lightweight tasks/ management system in a project — directory structure, tasks/CLAUDE.md procedures, global task manager, and main.md template.
SkillBFreshness: BClaude Code +9 - 0.518score
audit-rules
blakesims
Audits CLAUDE.md files to suggest high-signal, concise rules based on recent changes or a general audit.
SkillBFreshness: BClaude Code +9 - 0.513score
mockups
blakesims
create quick mockups of frontend design for fast iteration.
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Sep 18, 2026 (1 month ago)
- Last release
- Sep 14, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from blakesims
- 0.524score
cca-task-shape
blakesims
Shape a real task before execution.
SkillBFreshness: BClaude Code +9 - 0.524score
intent-harden
blakesims
Stress-test and tighten an Intent Contract before planning.
SkillBFreshness: BClaude Code +9 - 0.524score
doc-new-note
blakesims
Use when asked to create a "new note", "daily summary", "worklog entry", "weekly summary", or similar, in a project that has a worklog system…
SkillBFreshness: BClaude Code +9 - 0.521score
doc-init
blakesims
Set up a lightweight project documentation system — worklog (transient notes) plus refs (curated docs).
SkillBFreshness: BClaude Code +9 - 0.521score
doc-sweep
blakesims
End-of-day synthesis — reads today's worklog notes, resolves cross-note tensions, and fills today's daily summary.
SkillBFreshness: BClaude Code +9 - 0.521score
investigation-review
blakesims
Independently review a debugging conclusion against risk-critical evidence; return PASS or REVISE with concrete numbered blockers, separate from suggestions.
SkillBFreshness: BClaude 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 task-workflow 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
task-workflow skill. Agent Mart. https://agentmart.dev/skills/blakesims-task-workflow (last updated 2026-10-24)