Install any skill in seconds. Free to start, no credit card required.
Get Started Free →davekilleen/goal-backlog
Groom the open task pool goal by goal: see everything under each quarter goal, confirm doubtful goal links, retire what has gone stale, and mark what gets picked up first when that goal earns week-time.
Passed 11 of 18 test cases with the skill, and 3 of 18 without it.
and 9 more in the full report
10 fixed4 passed either way8 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
Earlier in this grooming session I declined your offer to convert recovered provisional goals into structured goal page shapes. Now that we are reviewing Goal 3, should you prompt me again to structure provisional goals?
| Checked | Without skill | With skill |
|---|---|---|
| Does not repeat the provisional goal offer after it was declined in the session. | Missed | Passed |
The instructions your model reads when the skill loads.
Turn a swampy task list back into a groomed pool: one goal at a time, decide what's still real, what's next, and what can go — so /week-plan pulls from decisions you already made instead of a flat dump.
/goal-backlog — Walk every group, one goal at a time/goal-backlog Q3-2026-goal-2 — Groom one goal onlyNever delete or move a task silently. Every exit from the backlog is the user's decision, and before any line is removed or relocated you show the exact lines that will change. Completing a task through the Work MCP is the only exit that needs no preview, because nothing is removed.
Call get_goal_backlog(goal_id="all") (or the single goal ID the user named).
Handle the response per CLAUDE.md's feature_status convention:
off (Quarter Goals room off): one calm line with the returned user_message, no error tone, and stop — there is nothing to groom without goals.not_installed / broken: surface the returned user_message and its fix path, then stop.unknown, or the tool errors: say the backlog could not be checked. Do not improvise a grooming pass from grep.If the pool is empty (open_task_count is 0), say so plainly and stop — no invented work.
Then show a one-screen overview in plain language, largest and stalest groups first:
"27 open tasks. Goal 2 (Launch mobile beta) holds 14 — 9 untouched for 3+ weeks. Goal 1 holds 6, all fresh. 5 tasks serve a pillar but no goal, and 2 are linked to nothing at all. Where do you want to start?"
Staleness comes from each task's staleness_days (age since the task was created); a group's stale_count uses the 3-week line the tool reports. Say "sat untouched for 3+ weeks" — never dress the number up as activity data it isn't.
For the chosen group, show: the count, the stale flags, tentative (?) links, and the current next-up order if one exists. Keep it to one screen; titles and ages, not raw metadata.
Work through three passes, in this order:
For each task marked tentative, ask one short question — "Does 'Draft partner FAQ' really serve this goal?" — and call confirm_goal_link(task_id, action="confirm") or action="clear". Cleared tasks fall back to their pillar group; mention that so they don't seem to vanish.
test cases passed, vs 3 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.