subagent-reports
/subagent-reports
Dodaje pisemny raport do każdego zadania delegowanego podagentom i zapisuje go w projekcie razem z wynikami.
- Autor
- hadong1502 ↗
- Repozytorium
- learning-skills ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 31.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 1
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 31.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 1 skrypt w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/log_subagent_report.py
Co robi i kiedy
Dodaje pisemny raport do każdego zadania delegowanego podagentom i zapisuje go w projekcie razem z wynikami.
Oryginalny opis (EN)
Makes a written report part of every subagent's deliverable and logs it in the project. Use EVERY time before delegating work to a subagent — the Agent tool (general-purpose, Explore, Plan, claude-code-guide, forks, custom agents), Workflow scripts, or parallel research/review agents — and again when the agent's result comes back. Also use when the user asks where a subagent's findings were saved, or to backfill a report for an agent that already ran.
Jak działa
- Before spawning: add the deliverable to the brief
- When the result comes back
- Safety net
Podgląd SKILL.md
--- name: subagent-reports description: >- Makes a written report part of every subagent's deliverable and logs it in the project. Use EVERY time before delegating work to a subagent — the Agent tool (general-purpose, Explore, Plan, claude-code-guide, forks, custom agents), Workflow scripts, or parallel research/review agents — and again when the agent's result comes back. Also use when the user asks where a subagent's findings were saved, or to backfill a report for an agent that already ran. --- # Subagent reports The user wants every subagent's work kept as a file, not just relayed in chat. Reports live in the project the session works in: ``` <project>/.claude/agent-reports/ INDEX.md one line per report (curated and raw) <YYYY-MM-DD>_<slug>.md curated report (this skill) _raw/<stamp>_<type>_<id8>.md automatic safety net (SubagentStop hook) ``` `<project>` is the session's working directory. `<slug>` is 3–6 kebab-case words naming the task (`course-ingestion-prior-art`). If the name exists, add `-2`, `-3`. ## Before spawning: add the deliverable to the brief Append the block below to every agent prompt, with the path filled in. In a Workflow script, append it to each `agent()` prompt, with one path per agent. **Agents that can write files** (general-purpose, claude, fork, and most custom agents) write the report themselves: ``` ## Report deliverable (required) Write your full report to: <absolute path to .claude/agent-reports/<date>_<slug>.md> Use the structure in ~/.claude/skills/subagent-reports/template.md. The file is the record: put every finding, source link, number and caveat in it, not just a summary. Then hand back the path plus a 3-6 bullet summary. ``` **Read-only agents** (Explore, Plan, claude-code-guide, anything without Write) can't save files. Instead, ask them to return the full report in the template's structure, and write it yourself afterwards. Never ask a read-only agent to write a file. ## When the result comes back 1. Open the report file and check that it is the full report, not just the summary. If it is missing or thin, write it yourself from the hand-back. Copy the hand-back word for word and add the frontmatter. 2. Add a line to `INDEX.md`: `- <date> · <status> · [<title>](<file>) · <agent type>` 3. Name the report path in your reply to the user, as a link. 4. Treat the report as the agent's claims. Check it before you rely on it, and record what you verified or corrected under `## Parent notes` in the report. Files the agent made outside the reports folder (prototypes, scripts, outputs) go under the report's "Files" section. If they're in the scratchpad or another temporary place, copy them next to the report into `<date>_<slug>/`, so the report doesn't point at files that will disappear. ## Safety net A `SubagentStop` hook in `~/.claude/settings.json` runs `scripts/log_subagent_report.py` every time a subagent finishes. It writes the brief and the final hand-back to `_raw/` and indexes it. That copy is automatic and unedited, and it doesn't replace the curated report. To backfill an agent that ran before the hook existed: ```bash python -I ~/.claude/skills/subagent-reports/scripts/log_subagent_report.py --transcript ~/.claude/projects/<project-key>/<session-id>/subagents/agent-<id>.jsonl --cwd <project> ``` The task `.output` files in the temp folder can be empty on Windows. The transcripts under `~/.claude/projects/.../subagents/` are the real record.
Pliki w paczce (3)
- SKILL.md 3 KB
- template.md 1 KB
- scripts/log_subagent_report.py 7 KB
Wersje
W bazie nie ma innych skilli o nazwie subagent-reports.
Migawka bazy z 13.11.2026 · wersja paczki 3223482d3958 (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
- 100subagent-best-practices
Opisuje, jak poprawnie definiować subagenty Claude Code w plikach agents/*.md: frontmatter, dobór narzędzi i przekazywanie kontekstu. - 100delegating-work
Pomaga zdecydować, kto wykona zadanie: ty, subagent czy inny agent z kontekstem, i jak go zbriefować w zależności od modelu. - 100create-subagent
Generuje jednofunkcyjne subagenty Claude Code do użycia przez narzędzie Task, z definicjami agentów, diagnostyką i konfiguracją uprawnień. - 95subagent-driven-development
Realizuje plan wdrożenia, przekazując niezależne zadania podagentom w jednej sesji roboczej. - 98agent-types
Dobiera typ agenta do zadania: wbudowane Bash, Explore, Plan i general-purpose oraz agenty z wtyczek do przeglądu i refaktoryzacji. - 100sub-agent-delegation
Deleguje pracę podagentom w Claude Agent SDK: definiuje ich, ogranicza narzędzia i ustala limity głębokości, równoległości i wydatków.