--- name: "delegating_work_to_humans" description: "Delegate work to staff, contractors, or a virtual assistant so it stays delegated: audit a week of tasks, score each by only-you factor, teachability, risk, and time, set a decision level per task, hand over with a pack covering outcome, standard, and check-in rhythm, then run a four week ramp with exit tests. Use when someone wants to hand off recurring work, onboard an assistant, stop taking delegated work back, or decide what only they should keep doing." --- # Purpose Most delegation fails in the follow-through, not the handoff. The work was vague, the decision limits were never named, or review only started after mistakes spread, so every task routes back to the delegator. This skill runs the full system: audit, score, set decision rights, hand over with a written pack, ramp over four weeks, and loosen control only where the output earns it. Bundled tools: `scripts/delegation_scorer.py` scores an audit CSV, `templates/handoff-pack.md` is the handover document, `templates/audit-sample.csv` is a worked input, `references/decision-rights.md` holds the level definitions, and `references/worked-example-four-weeks.md` shows one full role handover. # Workflow ## 1. Audit one real week - Log every recurring task for five working days as it happens: task, minutes spent, how many times it ran. Memory undercounts admin work, so log live, not from recall at the end of the week. - For each task, rate four things. Only-you from 0 to 5, where 5 means a legal duty, a signature, or judgment nobody else holds. Teachability from 1 to 5, where 5 means a checklist teaches it in under an hour. Risk from 1 to 5, where 5 means a mistake is costly, public, or hard to reverse. Value from 1 to 5, how good a use of the delegator time the task is. - Sort with the Eisenhower test as a first pass: urgent but not important work is the natural delegate pile, and work that is neither urgent nor important is a drop candidate, not a delegation candidate. ## 2. Score and sort - Put the audit in CSV with the header from `templates/audit-sample.csv` and run `python3 scripts/delegation_scorer.py `. - Read the verdicts. KEEP means only-you is 4 or higher, the task stays. KEEP_AND_TRAIN means the risk is high and teachability is low, so the assignee shadows before touching it. DELEGATE_NOW and DELEGATE_WITH_CHECKS are the handover pile, in that order. CONSIDER_DROPPING means the task is small, rare, and low value, so handing it over costs more than killing it. HOLD means do it once more yourself while recording how, then rescore. - The score is a sorting aid built for this skill, not a researched fact. Where the score and your knowledge of the person disagree, trust the knowledge and note why on the audit. ## 3. Set decision rights per task - Give every delegated task a level from 1 to 7, using `references/decision-rights.md`. Tell, Sell, Consult, Agree, Advise, Inquire, Delegate. The level applies to the decision area, so invoicing up to a limit can sit at 5 while refunds sit at 3. - Write the limits as numbers and named cases, not feelings. A spend limit, a list of senders or clients who always need approval, and a red list of calls that are never the assignee's: signing, deleting records, speaking for the business in a dispute. - Where several people touch one task, add RACI. One Responsible who does it, exactly one Accountable who owns the outcome, Consulted before, Informed after. Delegation moves the doing. The delegator usually stays Accountable, because handing over a task does not hand over answering for the result. ## 4. Hand over with a pack, not a chat message - Fill one `templates/handoff-pack.md` per task: outcome and definition of done, a good example, the common mistake, decision level and limits, inputs and access, trigger, deadline with timezone, checkpoints, check-in format, escalation path, and a backup approver. - Grant access as the assignee's own accounts with the least access the task needs. Never hand over the delegator login. - Close the handoff by having the assignee repeat back the outcome, the deadline, the decision limits, and the first risk they see. Correct gaps now. A repeat-back that surfaces a misunderstanding is the cheapest fix in the whole system. - Write the missing information rule into every pack: when a required input is absent, the assignee follows the stated fallback or asks, and never guesses. ## 5. Run the four week ramp - Week 1: hand over only the top two or three DELEGATE_NOW tasks. Review every delivery the same day. Short daily check-in. Assignee logs questions in a shared doc instead of guessing. - Week 2: add the next task only if week 1 passed its exit test, for example a full clean day with no corrections. Review moves to a daily batch. Start a rework log: every fix the delegator still makes, counted by task. - Week 3: add judgment work scored DELEGATE_WITH_CHECKS, with the delegator keeping final sign-off on the parts the audit marked high risk. - Week 4: fold the packs and every logged correction into standing procedures. Review four numbers with the assignee: time from assignment to start, clarification questions per task, tasks done by the original deadline, and rework per task. The full worked version is in `references/worked-example-four-weeks.md`. ## 6. Loosen, hold, or pull back - Raise one decision level for one area after the last few cycles needed no rescue and the assignee can state the standard back unprompted. Move review to spot checks, roughly one delivery in five. - Hold the level and fix the pack when rework stays high. Most repeat errors are a missing example, a vague definition of done, or an unnamed decision limit, and all three are pack fixes, not people fixes. - Drop one level after a real miss, say why out loud, and name what clean cycles earn it back. Never take a task back silently or redo the assignee work without telling them. Both teach people to hide problems and wait you out. # Output Contract When running a delegation plan, produce: 1. The scored audit table with a verdict per task and the recoverable hours per week. 2. A decision rights list: one level per delegated decision area, the numeric limits, and the red list. 3. One filled handoff pack per task being handed over in the current ramp week. 4. The ramp plan: this week scope, review cadence, exit test, and the date the level gets reviewed. 5. At each review, the four numbers (start lag, questions per task, on-time rate, rework per task) and the level change they justify, if any. # Operating Rules 1. Never delegate a task you cannot describe a good result for. Do it yourself once more while recording the steps, then delegate the recording. 2. Never delegate accountability. Tasks, decisions, and spending authority move. Answering for the outcome stays with the delegator. 3. Start narrow. Two tasks handed over cleanly beat ten handed over in a dump that all come back. 4. Review the output, not the status. A "done" message is not evidence. Open the deliverable. 5. Least privilege access, always, with the assignee on their own accounts. Broaden access task by task as the ramp earns it. 6. No fake precision. The scorer weights are a practical heuristic. Present them that way, and never cite delegation statistics you have not verified from a primary source. 7. Respect the person. Delegation that sticks gives real authority inside clear limits, credit for clean work, and a stated path to more. Handing someone only the work nobody wants, with no authority attached, is dumping, and it does not hold.