si
/si
Implementuje lub wznawia zadanie na podstawie istniejącego planu i kryteriów akceptacji, z testami opartymi na zachowaniu i zapisaną weryfikacją.
- Autor
- alexandrbasis ↗
- Repozytorium
- claudops ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 08.09.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 08.09.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Implementuje lub wznawia zadanie na podstawie istniejącego planu i kryteriów akceptacji, z testami opartymi na zachowaniu i zapisaną weryfikacją.
Oryginalny opis (EN)
Implement or resume a task from its existing plan and acceptance criteria, with behavior-first tests and recorded verification. Use for requested implementation; unresolved discovery belongs to /nf, planning to /ct, and independent review to /sr.
Jak działa
- 1. Read the active task and current state
- 2. Implement in verified slices
- 3. Verify the complete change
- 4. Save status and hand off
Podgląd SKILL.md
--- name: si description: Implement or resume a task from its existing plan and acceptance criteria, with behavior-first tests and recorded verification. Use for requested implementation; unresolved discovery belongs to /nf, planning to /ct, and independent review to /sr. allowed-tools: Agent AskUserQuestion Edit Read Write Bash Glob Grep Skill TodoWrite metadata: claude_argument_hint: '["task-path | technical-plan-path"]' compatibility: Portable clients may not enforce Claude Code invocation guards or project setup behavior. --- <!-- claudops-build: project-config-pointer --> > If `.claude/skills/si/SKILL.md.disabled` exists, stop before reading a fallback. > **Project configuration:** If the current project contains `.claude/skills/si/SKILL.md`, read and apply it instead of this bundled default. The project copy is the capability source of truth. > **Repository context:** Read `../setup/references/task-context.md` when resolving a task, project commands, named agent roles, or legacy `.claude/` resource paths. Use repository evidence and applicable project instructions; a missing local workflow copy does not require setup. Resolve bundled resources from the installed skill, never from the target cwd. # Implement a task Deliver the agreed outcome and leave verified progress that a person or fresh agent can resume. A short plan in the existing task record is sufficient when the work is clear. ## 1. Read the active task and current state Resolve the task with [shared task context](../setup/references/task-context.md), following its active plan or selected phase. Create the minimum record only when none exists. Read the agreed behavior, constraints, decisions, unfinished steps and relevant code/tests. Reuse findings that still match the code; inspect what is missing or changed. Fill structural gaps from the authorized request and repository evidence. Missing headings or a particular decomposition filename are not blockers. Resolve questions that materially change scope, behavior, contracts or safety before dependent implementation; continue independent work. Use /nf or /ct only when actual discovery or planning is needed. Inspect the working tree and preserve unrelated changes. Keep the selected checkout unless isolation helps. Reuse authorization for the same operations and scope; implementation alone does not authorize publication or tracker writes. Record the task as in progress using its existing status vocabulary. ## 2. Implement in verified slices For behavior changes, follow [TDD](../tdd/SKILL.md): observe a test fail for the intended reason, implement that behavior, then refactor within scope. Use the agreed testing strategy and real repository commands. Documentation, formatting and other non-behavior changes use appropriate validation. On resume, check current code and evidence; preserve existing work and report missing test chronology instead of rewriting history to manufacture RED. Implement the full agreed scope in dependency order. Resolve routine technical choices from repository patterns and domain terms. Record consequential deviations and their rationale once in the existing task; obtain a decision for changes to agreed scope or acceptance. For behavior that mutates application data or involves async interactions, read [mutation and async checks](references/implementation-checklists.md). For enumerated, visual or cross-surface requirements, read [verification recipes](references/verification-gate.md). Read the references that apply, using their checks for the behavior the task requires. When independent work has disjoint ownership and delegation offers a useful benefit, use [parallelization](../parallelization/SKILL.md). Otherwise work directly. The orchestrator owns shared task updates and verifies the combined result. After each meaningful slice, update its existing acceptance/checklist entry with observed evidence and the next unfinished action. If no entry exists, add a compact note in the task's plan or progress section. Keep one mapping; existing IDs and plain-language acceptance are both valid. Save progress before a pause so resume starts from these facts. ## 3. Verify the complete change Remove residue introduced by this task and run applicable project checks from their owning packages. Record command, working directory, result and the relevant revision or artifact. Reuse evidence only while the affected code and inputs remain unchanged; rerun affected checks after edits, cleanup or integration. A skipped or blocked check is not a pass. Compare every agreed acceptance item with observed evidence, including required integration or visible behavior. Code presence alone is insufficient proof of an outcome. Resolve failures and missing required checks before claiming implementation complete; a partial result may be handed over as a draft with its blockers named. ## 4. Save status and hand off Update the existing current status and next action in place, with links to code, verification, material deviations and remaining work. Keep old states as history. A separate completion report is unnecessary when these facts already exist; a finished phase leaves its parent open until the remaining phases are complete. Preserve links to completed phases, distinguish pending work from the selected phase, and advance only within the authorized scope. When review is requested or required, hand /sr the active task and actual diff/commit/PR target; a local diff is valid input. Continue already authorized review or delivery, read back actions you claim, and distinguish implementation, review, merge and deployment in the final result.
Pliki w paczce (3)
- SKILL.md 6 KB
- references/implementation-checklists.md 2 KB
- references/verification-gate.md 2 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · alexandrbasis/claudops | 85 | MIT |
| wszqkzqk/scrinium | 100 | GPL-3.0 |
| alexandrbasis/claudops | 85 | MIT |
Identyczna treść występuje jeszcze w 1 innych folderach na GitHubie. Tu pokazujemy jedną, najlepiej ocenioną kopię.
Migawka bazy z 28.10.2026 · wersja paczki 294772298c5e (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
- 100implement-issue
Realizuje zgłoszenie GitHub metodą tests-first: pisze testy na podstawie kryteriów akceptacji, otwiera roboczy PR i implementuje zmiany w commitach, aż testy przejdą. - 100writing-acceptance-tests
Pisze testy niezależne od kodu produkcyjnego, wyprowadzone ze specyfikacji, interfejsu publicznego lub kryteriów akceptacji, zanim powstanie implementacja. - 100subagent-driven-development
Wykonuje plan wdrożenia w bieżącej sesji, zadanie po zadaniu; do ewidencji kryteriów akceptacji służy inny silnik. - 100verify
Weryfikuje ukończoną funkcję względem kryteriów akceptacji, w tym testami przypadków granicznych. - 100rpi-plan
Tworzy lub wznawia plan wdrożenia oparty na dowodach, opcjonalnie z niezależną krytyką, gdy zadanie wymaga planu implementacji albo istniejący plan trzeba poprawić. - 10002-implement
Pisze kod według istniejącego planu, faza po fazie, aż spełnione zostaną wszystkie kryteria akceptacji.