fsa-announcements

Files (20)

• NOTICES.md
• SKILL.md
• SKILL_ja.md
• pytest.ini
• fields.md · references/
• sources.md · references/
• fetch_announcements.py · scripts/
• pyproject.toml · scripts/
• uv.lock · scripts/
• conftest.py · tests/
• news_rss.xml · tests/fixtures/
• recent_dec2025_excerpt.html · tests/fixtures/
• recent_jun2026_excerpt.html · tests/fixtures/
• recent_two_months_2026.html · tests/fixtures/
• test_cli.py · tests/
• test_filter_and_merge.py · tests/
• test_http_get.py · tests/
• test_integration.py · tests/
• test_recent.py · tests/
• test_rss.py · tests/

SKILL.md (first lines)

# FSA Announcements

Aggregate and filter announcements published by the Financial Services
Agency of Japan (FSA) — combining the official news RSS feed with the
yearly `recent{YYYY}.html` archive pages.

## When to use this skill

Use this skill when the user wants to:

- Monitor recent FSA **press releases** (報道発表), **council outcomes**
  (審議会等), or **international relations** announcements (国際関係)
- Build a date-bounded list of FSA announcements within an arbitrary
  window (today back to 2021) — for instance, "all 報道発表 from
  2026-04-01 to today"
- Filter by one or more categories (e.g. only 報道発表 and 審議会等)
- Cross-reference an item observed elsewhere against the FSA's own list
  to confirm publication date and category

This skill is complementary to `boj-statistics` (BOJ time-series data)
and `jp-security-advisories` (JPCERT/MyJVN security disclosures).

## Sources

| Source | Backend | Coverage |
|---|---|---|
| News RSS | `https://www.fsa.go.jp/fsaNewsListAll_rss2.xml` | The newest ~15 items across all categories. RFC 822 RSS 2.0; `pubDate` uses the non-standard `JST` zone abbreviation, normalised by this skill. |
| `recent{YYYY}.html` | `https://www.fsa.go.jp/recent/recent{YYYY}.html` | One HTML page per Western year, with month-by-month `<ul>` blocks. Pages exist from 2021 onward. |

The RSS feed is small but always current; the recent pages are large
(typically 1,100+ entries per full year) and provide authoritative
category labels. When both sources contribute to a date range, the
skill de-duplicates by URL and prefers the recent-page entry because it
carries an authoritative category derived from the FSA's own label
markup.

See [references/sources.md](references/sources.md) for endpoint details
and known limits.

## Workflow

### 1. Determine arguments

| Argument | Description | Default |
|---|---|---|
| `--start-date` | Inclusive start date in `YYYY-MM-DD` (JST calendar). **Required.** | — |
| `--end-date` | Inclusive end date. | today (JST) |
| `--source` | One of `rss`, `recent`, `both` | `both` |
| `--category` | Filter by category (repeatable). One of: `報道発表`, `記者会見`, `審議会等`, `国際関係`, `イベント`, `刊行物`, `採用`, `調達`, `その他` | (none → all categories) |
| `--cache-dir` | If supplied, cache HTTP responses under this directory using ETag / Last-Modified conditional GETs. **Omitted by default — no files are written when the flag is absent.** | (off) |
| `--diagnose` | Print per-source counts to stderr (helpful when investigating coverage anomalies or FSA HTML changes) | off |
| `--log-level` | `DEBUG`/`INFO`/`WARNING`/`ERROR` (or via `FSA_LOG_LEVEL`) | `WARNING` |

If the user's intent is ambiguous (no date hint, no keyword), confirm
the date range before proceeding.

### 2. Run

```bash
uv run --project scripts scripts/fetch_announcements.py \
  --start-date <YYYY-MM-DD> [--end-date <YYYY-MM-DD>] \
  [--source rss|recent|both] \
  [--category 報道発表 ...] \
  [--cache-dir <PATH>] [--diagnose]
```

The script writes a JSON array to stdout, one object per announcement,
sorted by `published` descending. See [references/fields.md](references/fields.md)
for the field schema.

### 3. Summarise

Either:

- Pass the JSON straight to the user when they asked for a list,
- Or summarise — typically grouping by category and date range,
  highlighting items the user is likely to care about (e.g. supervisory
  guideline drafts inside 報道発表), and citing the FSA URL alongside.

### 4. Handle large windows in a sub-agent

A multi-year `recent{YYYY}.html` pull can return thousands of items.
Run such pulls inside a sub-agent so that the parent context is not
flooded with raw entries.

## Cache behaviour (important for restricted environments)

By default the script writes **no files** to the local filesystem.
Pass `--cache-dir <path>` to opt into HTTP caching:

- Each response is stored as `{sha256(url)}.body` and `{sha256(url)}.meta.json`
- Subsequent runs send `If-None-Match` / `If-Modified-Since` and reuse
  the cached body on `304 Not Modified`
- Errors during caching are non-fatal; the network response always
  takes precedence

This default suits financial institutions where third-party scripts
writing to local disk is typically not permitted.

## Script details

- **Runtime:** Python 3.11+, standard library only, run via `uv run --project scripts`
- **Logging:** `logging` module to stderr (`--log-level` or `FSA_LOG_LEVEL`)
- **Retries:** Exponential backoff, max 3 attempts per HTTP request
- **Output:** JSON to stdout; errors to stderr with non-zero exit code
- **Network access:** `www.fsa.go.jp`

## Examples

Last 7 days, all categories:
```bash
uv run --project scripts scripts/fetch_announcements.py --start-date 2026-06-01
```

Press releases only, full quarter:
```bash
uv run --project scripts scripts/fetch_announcements.py \
  --start-date 2026-04-01 --end-date 2026-06-30 \
  --category 報道発表

From aws-samples/sample-fsi-reference-architecture-jp, MIT-0.

More from aws-samples/sample-fsi-reference-architecture-jp