dsh-sekaisync-connect

Configuration

Precedence: environment variables > profile row config > SEKAISYNC_CONFIG file > plugin-directory config.local.json > config.json > automatic discovery.

Setting	Description
SEKAISYNC_STORE	Store directory containing kb/. When omitted, discovery checks /store, cwd/store, and the repository store located with python -c "import sekaisync"
SEKAISYNC_ROOT	SekaiSync repository root, used as the working directory for python -m sekaisync
SEKAISYNC_PYTHON	Python executable; defaults to python
SEKAISYNC_PORT	Port to probe for an existing server; defaults to 8787. If a ready SekaiSync service is already listening there, it is reused instead of starting another process. 0 retains its meaning: skip the external probe and start a managed service directly
SEKAISYNC_MAX_RESPONSE_BYTES	Byte budget for one successful HTTP response; defaults to 134217728 (128 MiB), configurable from 65536–1073741824. It can also be set as maxResponseBytes in a configuration file or profile row. Exceeding the budget cancels the read and returns an explicit error

You can also override these settings in your own profile patch; plugin upgrades do not overwrite that layer:

- id: sekaisync-connect
  name: 'dsh-sekaisync-connect'
  config:
    store: 'D:\\sekaisync\\store'
    python: 'py'

Keep the row id as sekaisync-connect. It is the key the panel writes through, so if you copy this snippet into a profile that still uses an older id, update that row's id to match — otherwise the panel and the saved path refer to different rows.

• HTTP connects only to loopback and rejects redirects. Health and error responses are limited to 64 KiB and 4 KiB respectively. The default successful-response budget is 128 MiB, covering typical body results with limit=100 and max_text_chars=200000. max_text_chars=0 still means full text upstream; if an exceptionally large full-text response exceeds the transfer budget, increase maxResponseBytes (up to 1 GiB) or read in batches.
• External-port reuse still follows the existing /health ready/status check. That endpoint carries no store identifier, so you must ensure the configured port serves the intended knowledge base.

Usage Recommendations (Token Discipline)

1. Use sekai_resolve before generating any localized text. Use sekai_fact for precise facts; do not let the model answer from memory.
2. For an unfamiliar term, an uncertain topic, or a character name with an unknown source, first use sekai_probe to determine whether it concerns Project Sekai, then route to local tools or the web. Matches directly suggest a next_tool.
3. For time-sensitive questions such as current events, gachas, and maintenance, check freshness and coverage with sekai_status first.
4. To verify a localized name in context, use sekai_term to confirm the term exists, then sekai_penetrate to inspect aligned lines across languages. Remember that missing explicitly means there is no corresponding line in that language; it is not an error.
5. sekai_web is slow; use it only when quoting original story text. Try sekai_lookup / sekai_term first, or use sekai_probe to confirm the direction.
6. Ask directly with event shorthand or an official event name, whether a unit-focused event or World Link. sekai_lookup automatically includes activity resolution, and sekai_alias supports shorthand, official event names, and natural-language questions.
7. If a result is truncated, reduce limit or use a more precise query.