Interactively authors stop-proof goal files for autonomous /goal runs.
npx skills add https://github.com/dxiiren/project-skeleton --skill define-goal-dxiiren
Please help me install this Agent Skill. Skill: define-goal Source: https://github.com/dxiiren/project-skeleton/tree/main/.claude/skills/define-goal Command: npx skills add https://github.com/dxiiren/project-skeleton --skill define-goal-dxiiren
{topic}-goal.md with a checkable STOP CONDITION, a fully enumerated work list with terminal statuses, banned "deferred" statuses, a code-grounded blocker definition, a durable JSONL state ledger, stall detection, a hard budget, and a resume protocol. - Handoff to autonomous execution: Prints the exact /goal invocation so a fresh instance runs the goal unattended, resuming from the ledger after crashes or context compaction. - Use Case: Before an overnight refactor sweep, run /define-goal refactor-logging to be interrogated about scope, guardrails, and budget, then receive a goal file that an unattended agent works through row by row until every item reaches a terminal status. ## Quick Start Ask the assistant to define a goal for an autonomous run, for example by saying "/define-goal audit-workflows", then answer its questions until you confirm the spec.High-intent search queries and answers about installing and using this skill.
Invoke /define-goal with a topic slug and answer the interrogation rounds covering mission, work list, definition of done, guardrails, budget, and concurrency. After you explicitly confirm the assembled spec, it writes {topic}-goal.md into .claude/checklist/{topic}/ and prints the /goal command to run it.
A stop-proof goal has a checkable STOP CONDITION, a fully enumerated work list with terminal statuses (DONE, GAP, BLOCKED), banned statuses like "deferred", a code-grounded blocker definition, and a resume protocol. These five pillars prevent the agent from writing an early summary and quitting with rows still open.
Yes. Every iteration appends one JSON line to a {topic}-state.jsonl ledger recording the completed item, command, and criteria results. A fresh session reads the ledger first and resumes by set difference against the work list, never by cursor, so concurrent or out-of-order completions are handled.
Run serially when items touch the same file, share a stateful resource like one server or database, or when one item's output feeds another. Parallel batches of 4-8 subagents are allowed only for genuinely independent items, and require the ledger to record an item per line for set-based resume.
The interrogation loop is the core value: a goal written from half-formed answers fails mid-run when nobody is available to clarify. The skill re-presents the full assembled spec after every correction and only writes after an explicit confirmation like "confirmed" or "ship it".
JunMA98Turn rough CS ideas into scoped research plans with milestones, risks, and next actions.
dxiirenInteractively authors stop-proof goal files for autonomous /goal runs.
hivgb1-aiValidates project feasibility by testing market reach and conversion before scoping.
FRIKKernCreate and dispatch tasks from natural-language goals using doey commands.