Trust, recovery, evidence

Protect secrets, constrain claims, and leave the campaign recoverable.

The bundled tools are useful precisely because their verdicts are narrow. They guard local state; they do not authenticate humans, understand semantic spoilers, certify table quality, or make publication safe.

Privacy and storage

Local tools, user-chosen files, host-dependent transmission.

Skill sourceMarkdown, JSON, images, and Python standard-library scripts in the installed directory.

Campaign dataWritten only to the workspace and output paths chosen by the operator.

NetworkThe bundled scripts make no network requests. The AI host may transmit context under its own configuration and terms.

ControlsNo encryption, authentication, account, telemetry, cloud database, or deletion API is included.

Campaign material may contain identities, schedules, private messages, unreleased plots, licensed text, accessibility needs, or consent boundaries. Minimize what enters the AI context and never publish an export without reading every word.

Security boundary

Imported material is data, not authority.

Evidence

Static files passed their checks. Live VTTs have not testified.

Package

The curated self-check passed for all 32 instrument cores and required export assets.

Ledger guard

Local tests reject invalid identity, links, visibility paths, asset digests, and unauthorized promotion.

Frozen capture

Local tests exercise campaign-root path confinement, link and reparse rejection, source drift detection, and downstream use of copied bytes.

Generic packs

Local tests exercise deterministic ZIPs, member digests, previews, player-candidate approval, mutation rejection, and unchanged final bytes.

Target adapters

Alchemy and Foundry bundles pass local shape, path, mapping, loss-report, audience-permission, and repeat-build checks. The Foundry importer passes JavaScript syntax checking.

Live targets

Alchemy import and Foundry 14.365 recognition, database creation, rendering, grid alignment, and repeat-run behavior are not tested.

Fresh-host installation, Codex or Claude discovery of this standalone checkout, live table fun, balance, rules accuracy, accessibility of generated play, and live VTT behavior are not independently verified. See the exact matrix and bound receipts in Provenance and verification records.

For target-specific limits and import steps, use the offline export guide.

Recovery

Preserve the failing state before fixing it.

Discovery failure

Confirm the directory name and direct SKILL.md placement, run self-check, restart the host, then use the fallback if discovery remains unverified.

Ledger failure

Snapshot or copy first. Repair the first explicit violation without exposing secrets, deleting authority, or silently discarding a conflicting claim.

Export concern

Preserve the candidate, preview, audit, and error. Correct the campaign source, choose a fresh output path, rebuild, and manually review the complete regenerated artifact before approving it.

python -B scripts/snapshot_campaign.py C:\Games\MyCampaign
python -B scripts/validate_ledger.py C:\Games\RestoredCampaign\campaign-ledger.json

A created archive is not a verified backup until it has been copied, hash-checked, restored elsewhere, and validated.

Update, remove, clean up

The skill and the campaign are separate custody domains.

Before updating a Git install, inspect local changes, then pull fast-forward only, rerun self-check, restart the host, and repeat discovery/invocation verification.

git -C "$env:USERPROFILE\.codex\skills\ludis-continuum" status --short
git -C "$env:USERPROFILE\.codex\skills\ludis-continuum" pull --ff-only
python -B "$env:USERPROFILE\.codex\skills\ludis-continuum\scripts\self_check.py"

To remove the skill, verify the exact directory and move it to a dated backup location before deletion. Restart the host and confirm it is no longer discoverable.

Removing the skill does not remove campaign ledgers, player exports, session artifacts, checkpoint ZIPs, host histories, Git clones, backups, or synced copies. Inventory and dispose of each according to participant consent, rights, and retention needs.

Accessibility

Documentation support is not a table-accessibility certificate.

The site provides skip links, landmarks, keyboard focus, responsive text, reduced-motion support, descriptive alternatives, labeled statuses, and copyable code. Hero art is supplementary; essential instructions remain text.

Ludis can help plan accessible handouts or ask about preferences, but humans must review generated material with the people who will use it. Read the accessibility statement.

Support and terms

Report the defect without leaking the campaign.

Use GitHub Issues for reproducible public defects. Include host/version, OS/Python, installation scope, exact observed stage, command and exit code, and a redacted synthetic fixture. For sensitive security reports, use GitHub's private vulnerability channel when available or request a secure contact without posting details.

Ludis Continuum is available under the MIT License. Contribution rules live in CONTRIBUTING.md.