dev10x:estimate
/dev10x:estimate
Szacuje czas realizacji zadania lub projektu, gdy implementację wykonują agenci AI, a przegląd i testy robią ludzie.
- Autor
- dev10x-guru ↗
- Repozytorium
- dev10x-claude ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 12.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 12.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Szacuje czas realizacji zadania lub projektu, gdy implementację wykonują agenci AI, a przegląd i testy robią ludzie.
Wyzwalacze z opisu
Oryginalny opis (EN)
Estimate how long a ticket or project will take when AI agents do the implementation and humans still gate review, testing and release — so a lead plans on numbers that match how the work actually ships, instead of one engineer's human-pace guess. TRIGGER when: sizing a ticket or project, filling an "estimated complexity" column, or answering "how long would it take to…?", "what would it take to…?", "when could this ship?" — inside or outside a scoping skill. DO NOT TRIGGER when: tracking time already spent (use a time tracker), or reporting what shipped (use dev10x:release-notes).
Jak działa
- Why this exists
- Method
- Output format
- Subagents
- Anti-patterns
Podgląd SKILL.md
--- name: dev10x:estimate description: > Estimate how long a ticket or project will take when AI agents do the implementation and humans still gate review, testing and release — so a lead plans on numbers that match how the work actually ships, instead of one engineer's human-pace guess. TRIGGER when: sizing a ticket or project, filling an "estimated complexity" column, or answering "how long would it take to…?", "what would it take to…?", "when could this ship?" — inside or outside a scoping skill. DO NOT TRIGGER when: tracking time already spent (use a time tracker), or reporting what shipped (use dev10x:release-notes). user-invocable: true invocation-name: dev10x:estimate allowed-tools: - Read - Grep - Glob - mcp__plugin_dev10x_cli__pr_list - mcp__plugin_dev10x_cli__pr_get --- # dev10x:estimate — Agent-Built Work Estimation **Announce:** "Using dev10x:estimate to size [ticket/project]." This is the single source of truth for estimation in Dev10x. `dev10x:project-scope`, `dev10x:ticket-scope` and `dev10x:work-on` delegate here, and the SessionStart guidance points every other estimate here too (GH-1495). ## Why this exists Claude's default estimate assumes one engineer at human pace and blends coding, testing and self-review into one number. It has no review wait, no CI time, and no docs. On agent-built work, that one number is wrong twice. It overshoots the mechanical work agents finish in hours. It also ignores the human gates, which dominate the calendar. Evidence: release R1 of a Codex port was estimated at "1-1.5 weeks, one engineer". Implementation, tests and live verification landed the same day, and the review and merge were never counted. ## Method ### 1. Size the change in points Points measure complexity and risk, not time. | Points | Complexity | Typical shape | |--------|------------|---------------| | 1 | Trivial | One file, existing pattern, no new tests | | 2 | Small | A few files, existing pattern | | 3 | Medium | New tests, one module boundary crossed | | 5 | Large | New pattern, or several modules | | 8 | Complex | Cross-layer, migration, or external dependency | | 13 | Epic-sized | **Split it** — do not estimate as one unit | Raise one step for any of these: a new pattern rather than an existing one, a database migration, an external dependency, or an unfamiliar codebase area. ### 2. Break out the five line items Every estimate reports all five. A line that does not apply is stated as `0` with the reason, never omitted. | Line item | What it covers | Effort (hands-on) | Elapsed (calendar) | |-----------|----------------|-------------------|--------------------| | **Agent implementation** | Agent coding, agent-run tests, agent self-review | Supervisor briefing + steering | Agent wall-clock | | **Human review** | Reading the diff, writing feedback, re-reading fixes | Reviewer hours | Same as effort | | **Review wait** | Queue time until a reviewer picks it up, per cycle | `0` — nobody works | Often the dominant term | | **Testing** | CI runs, manual/QA passes, live or end-to-end checks | QA and verification hours | CI wall-clock × cycles + QA | | **Documentation** | User docs, release notes, internal process docs | Writer hours | Same as effort | ### 3. Apply the default ranges, then calibrate Use these defaults only when the project has no history. Prefer calibration. List the last 15-20 merged PRs with `mcp__plugin_dev10x_cli__pr_list`, and call `pr_get` only on the few you need review timestamps from. Measure created-to-merged, review-requested-to-first-review and CI duration, then replace the defaults and say which you used. | Points | Agent elapsed | Supervisor effort | Review effort / cycle | Review cycles | |--------|---------------|-------------------|-----------------------|---------------| | 1 | < 30 min | 10 min | 10-15 min | 1 | | 2 | ~1 h | 15-30 min | 15-30 min | 1 | | 3 | 2-4 h | 30-60 min | 30-60 min | 1-2 | | 5 | 0.5-1 day | 1-2 h | 1-2 h | 2 | | 8 | 1-2 days | 2-4 h | 2-4 h | 2-3 | Default review wait: 0.5-1 business day per cycle on a team repo. On a solo-maintainer repo it is the supervisor's availability, often same-day. Default testing: the repo's CI duration × (review cycles + 1), plus 0-2 h of manual QA when the change is user-facing. Default documentation: 0 h internal-only, 0.5-2 h user-facing. ### 4. Aggregate effort and elapsed separately - **Effort** is the sum of hands-on hours across all five lines. - **Elapsed** follows the critical path. Agent work on independent tickets runs in parallel. Review, review wait and QA run in sequence per ticket, and a single reviewer serialises every ticket they own. - Never add review wait to effort. It adds calendar time and costs no one any work. ### 5. Project-level estimates Estimate each ticket with steps 1-4, then roll up: 1. Group tickets by milestone and by blocking chain. 2. Agent implementation of unblocked tickets overlaps. Take the longest chain, not the sum. 3. Human review does not overlap for one reviewer. Sum review effort per reviewer, and add one review wait per cycle along the chain. 4. Report per milestone and for the whole project. `dev10x:project-scope`'s "estimated complexity" column carries the points from step 1 plus the ticket's elapsed range. ## Output format ``` Estimate — <ticket or project> Points: <n> (<why>) | Line item | Effort | Elapsed | |----------------------|----------|----------| | Agent implementation | <h> | <h/d> | | Human review | <h> | <h> | | Review wait | 0 | <d> | | Testing | <h> | <h/d> | | Documentation | <h> | <h> | | **Total** | **<h>** | **<d>** | Basis: <calibrated from … | defaults>. Assumptions: <reviewer, CI time, QA need>. ``` ## Subagents SessionStart guidance does not reach a dispatched subagent, so the always-present pointer to this skill is absent there. A skill that asks a subagent for an estimate must inline this method or tell the subagent to Read this file — never assume it already knows it. ## Anti-patterns | Anti-pattern | Why it misleads | Instead | |--------------|-----------------|---------| | One number, "one engineer, 1-1.5 weeks" | Blends agent work with human gates and hides both | Five lines, effort and elapsed | | Agent coding at human pace | Overshoots mechanical work by days | Agent-elapsed column above | | Omitting review wait | Understates calendar time, usually the biggest term | State it, even as an assumption | | Summing parallel agent work | Overstates elapsed on multi-ticket plans | Critical path, step 4 | | Estimating a 13 | Too uncertain to plan on | Split, then estimate the parts |
Pliki w paczce (2)
- SKILL.md 7 KB
- evals/evals.json 6 KB
Wersje
W bazie nie ma innych skilli o nazwie dev10x:estimate.
Migawka bazy z 13.11.2026 · wersja paczki def0b455bc35 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100ticket
Prowadzi ticket z trackera przez analizę, wycenę, kod i przegląd, a człowiek zatwierdza wycenę i każdy zapis. - 96ticket-estimate
Szacuje zgłoszenie Jira w punktach historyjek i dzieli je na dwie osobne prace, gdy to w rzeczywistości dwie rzeczy. - 100wise-estimation
Szacuje pracę w punktach historyjek w skali Fibonacciego od 0,5 do 13 SP i wyjaśnia, jak wyceniać niepewne zadania. - 95automation-scoping
Szacuje koszt i czas automatyzacji zestawu przypadków testowych przed startem prac, w tym fundamentów frameworka i CI. - 100ticket-estimate
Ocenia nakład i ryzyko zgłoszenia Jira na podstawie podobnych, wcześniej zrealizowanych prac i historii incydentów. - 100eta
Szacuje, ile zajmie zadanie w Claude Code, rozdzielając czas agenta i użytkownika, z zakresem, największym ryzykiem i przyspieszeniami.