Przejdź do treści
Biblioteka skilli
Dane i analityka · Automatyzacja

weekly-report

/weekly-report

Buduje i publikuje raport dwa razy w tygodniu: siedem kanałów, warstwy daleka, średnia i bliska oraz synteza.

Repozytorium
semis-weekly ↗
Licencja
MIT
Ostatnia zmiana repozytorium
07.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 07.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Buduje i publikuje raport dwa razy w tygodniu: siedem kanałów, warstwy daleka, średnia i bliska oraz synteza.

Wyzwalacze z opisu

s report, or to rerun one channel (for example
Oryginalny opis (EN)

Build and publish the semis-weekly far-layer report (7 channels → far/mid/near → synthesis). Use when the user asks to run, build, update or publish this week's report, or to rerun one channel (for example "rerun channel C4"), in any language.

Jak działa

  1. 0. Pick the issue
  2. 1. Code
  3. 2. Level 1: seven channels, judged separately
  4. 3. Check and record level 1
  5. 4. Levels 2 and 3: written by the main model
  6. 5. Render and publish
  7. 6. Calibration
  8. Do not
Sekcje z nagłówków SKILL.md, w oryginale · 106 linii, 774 słów
Podgląd SKILL.md
---
name: weekly-report
description: Build and publish the semis-weekly far-layer report (7 channels → far/mid/near → synthesis). Use when the user asks to run, build, update or publish this week's report, or to rerun one channel (for example "rerun channel C4"), in any language.
---

# Build this week's far-layer report

Project root: the `semis-weekly/` directory that contains this skill. Python is `.venv/Scripts/python.exe`
(3.12, managed with uv); `P` below means that interpreter. Run every command from the project root.

Language: `SEMIS_LANG` in `.env` (default en), or pass `--lang` before the subcommand, e.g. `P -m semis --lang en prepare`.
Use the same language for every step of one issue. Below, `<lang>` is that language; its folder is `playbook/<lang>/`
(or `private/playbook/<lang>/` for a language kept out of the repo).

Rules live in `playbook/<lang>/playbook.md` (three-level aggregation, log formats, evidence tags). Read the files
each time; do not work from memory.

## 0. Pick the issue

The issue id is the ISO week, e.g. `2026-W44` (`semis signals` prints it on its first line). If the user gave
news or an FOMC result for this week, write it to `logs/<week>/inputs/notes.md` first (C7 reads it).

## 1. Code

```
P -m semis prepare
```

Fetches everything, computes signals (also written to signal history), builds the 7 input packets. Check the
output: for fetch failures or `na` signals, find the cause (`P -m semis status` lists unresolved failures). Do not
skip a failing source to get the report out; the report says "no data" instead. If code changed, run
`P -m unittest discover -s tests` first.

FOMC statements come out at 2pm New York time on the last meeting day and `prepare` fetches them. If you run
before that, C7 says the result is pending; once it is out, rerun `P -m semis fetch --channels C7`,
`P -m semis packet --channels C7 --run <week>`, the C7 level-1 log, then levels 2 and 3.

## 2. Level 1: seven channels, judged separately

If `logs/<week>/` already has a log for a channel (a rerun), move it to `logs/<week>/superseded/` and add the
reason to the file name.

Launch 7 Sonnet subagents in parallel (Agent tool, model: sonnet), one per channel. The whole task is:

> You are a level-1 channel analyst for the semis-weekly semiconductor report. First run this command in Bash;
> its output is your complete task. Follow it:
> `cd "<project root, posix path>" && ./.venv/Scripts/python.exe -m semis --lang <lang> prompt --run <week> --channel C<n>`

If the user only wants one channel rerun, launch only that one.

## 3. Check and record level 1

```
P -m semis check  --run <week>
P -m semis ingest --run <week>
```

When `check` lists internal IDs, the log must say the rule's name or meaning instead. When it lists numbers without a source, compare them with the input packet: if it is a unit conversion,
add it to `_SCALES`; if the model invented it, have that channel rewrite (SendMessage to its agent, or relaunch).
`ingest` runs `check` first and refuses to record on failure.

## 4. Levels 2 and 3: written by the main model

Follow the formats in `playbook/<lang>/aggregation.md`:

- `logs/<week>/near.md`, `mid.md`, `far.md`: each reads only the channel logs listed for it in section 1, answers
  that horizon's question, and checks the transmission chains CH-A to CH-D.
- `logs/<week>/synthesis.md`: reads only the three horizon logs and `playbook/<lang>/invariants.md`, in this order:
  roles, exception checks, invariant check, action.

Write them following `playbook/<lang>/style.md`. Numbers come only from the logs one level down. Rule and
signal IDs go only into the json `rule_refs` and `signal_refs`; the text uses the names in `playbook/<lang>/glossary.yaml`,
and `check` rejects IDs anywhere a reader would see them. Any departure from a rule
goes under "Deviations and uncertainty" and into the json `deviation`. If a rule itself is wrong (as CAL-03 was
in 2026-W44), fix the playbook in every language folder you use and add a change-log row.

Then run `check` and `ingest` again.

## 5. Render and publish

```
P -m semis render --run <week>
```

Read `reports/artifacts.json`. If this issue already has a URL, publish `reports/<week>.html` with the Artifact
tool and pass that `url`. Otherwise publish a new artifact (icon: `report`) and write the new URL back to
`artifacts.json`.

Reply with the link and one paragraph: the synthesis action, the three horizon states, and the one transmission
chain or signal most worth watching.

## 6. Calibration

```
P -m semis review
```

Lists judgments from 4–8 weeks ago with no recorded outcome. If there are any, list them for the user, ask
which held, then:

```
P -m semis calibrate --run <week> --item 2:near:state --outcome right|wrong|unclear --note "..."
```

## Do not

- Change thresholds in `signals/signals.yaml` without the user's approval. Changes go in its change log.
- Write the author's words or link to their posts. Use only the paraphrases already in the playbook.
- Put the contact email anywhere except the SEC User-Agent (`.env`).

Wersje

RepozytoriumWynikLicencja
Ta strona · Jackmeson1/semis-weekly100MIT
corgicorner/bisibility100AGPL-3.0
hayat-nextoriainfotech/social-media-skills100MIT
logly/mureo100Apache-2.0
calven-ai/marketing-as-code100MIT
lingegrace/claude-research-skills100MIT
fabiozagaria/learning-ledger100MIT
zhbehx/weekly-report100MIT
xuanyuanli/skills100MIT

Migawka bazy z 13.11.2026 · wersja paczki 223269eb733f (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

  • 100ga4-weekly-report
    Buduje tygodniowy lub miesięczny raport GA4 ze zmianami tydzień do tygodnia i rok do roku oraz oceną, czy wahania są realne.
  • 100weekly-synthesis
    Przygotowuje tygodniową syntezę lub raport w formie zestawienia tego, co trzeba wiedzieć w tym tygodniu, na podstawie raportów zespołu.
  • 100weekly-update
    Pisze tygodniowy raport zespołu z zapisu tygodnia, dla zespołu, menedżera, klienta lub zarządu, w formie podsumowania.
  • 86weekly-briefing
    Zamienia tygodniowe liczby w krótkie podsumowanie prostym językiem i wskazuje do trzech spraw na następny tydzień.
  • 90channel-report
    Pisze prosty raport wyników kanałów społecznościowych podłączonych do Postra, z liczbami za okres i opisem zmian.
  • 100web-analyst
    Przygotowuje tygodniowy raport o stronie: sesje według kanału i źródła, pobrania przez silniki, strony docelowe, CTA i konwersje, przyczynę zmian oraz jedną rzecz do zrobienia.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.