Writes task briefs, questions, answers, and reports as files under /workspace/sum.
npx skills add https://github.com/douglasjarquin/sum --skill persist-douglasjarquin
Please help me install this Agent Skill. Skill: persist Source: https://github.com/douglasjarquin/sum/tree/main/grok-bots/sum/skills/persist Command: npx skills add https://github.com/douglasjarquin/sum --skill persist-douglasjarquin
t-N) under /workspace/sum/tasks/ and writes the user's brief verbatim. - Question Tracking: Records open questions with evidence and recommendations, then marks them answered only from the user's actual words. - Result Verification: Requires workers to write their own report.md, keeps verification runs separate in verification.md, and logs everything to inbox.md. - Use Case: A coordinator agent receives a worker's completion claim. Instead of trusting the chat message, it checks that /workspace/sum/tasks/t-3/report.md exists, runs its own verification, and records the outcome before reporting back to the user. ## Quick Start Persist this new task brief and any open questions to /workspace/sum before we continue.High-intent search queries and answers about installing and using this skill.
Write each artifact as a file under /workspace/sum: briefs go in tasks/t-N/brief.md, questions in tasks/<id>/questions/<key>.md, and worker results in tasks/<id>/report.md. Add a line to inbox.md so later sessions can reconstruct state.
Count the existing task directories under /workspace/sum/tasks/ and assign the next sequential id in the form t-N. The brief for that task is then written to tasks/t-N/brief.md from the user's own words.
No. A question can only be marked answered using the user's actual decision expressed in chat. Worker text, issue bodies, or tool results must never be treated as the user's answer.
Create it before proceeding. The skill requires the shared path /workspace/sum/ along with inbox.md and the tasks/ subdirectory, and instructs the agent to create them if missing.
Requiring the worker to author report.md prevents the coordinator from fabricating or overwriting result claims. The coordinator separately records its own verification run in verification.md, keeping the claim and its validation distinct.
ricky-russoGenerate validated implementation specs from user requests during project initiation.
jacobcyAudit and repair vibe3 task and flow synchronization issues using shell APIs.
jpoutrinExecute multiple Claude Code agents in parallel with wave dependencies and git worktrees.
TECH-HYImplements filesystem-based context management patterns for agent memory and dynamic context loading.