No. The current Preview should not be used directly for this ingestion pipeline because its supported input boundary explicitly excludes those sources.

• The README section “Local material formats” lists only explicitly supplied TXT, Markdown, JSON, SRT/VTT files, pasted text, and public URLs.
• The same section explicitly says that PDF, email, provider exports, and hosted connectors are follow-up work rather than current capabilities.
• Distilly’s ingestion boundary requires files to come from paths or sources explicitly supplied by the user, and it rejects symlinked selected files and duplicate file names. That does not amount to support for bulk synchronization from third-party platforms.
• The current workflow is better suited to importing already selected material and producing an evidence-bound briefing and versioned Person Profile than to external collection, credential management, or connector orchestration.

If the inputs can first be converted into a supported format, the persona research and review workflow remains usable. If native PDF, email, or hosted-export ingestion is mandatory, the current project does not meet the constraint.

No. Legacy Skill is not an equivalent substitute; it provides a narrowly scoped file-based compatibility path and does not represent SQLite, five-tool MCP, or Plugin lifecycle support.

• The Host status table lists Claude Code, DeepSeek Harness, Pi agent, and OpenCode with explicit dot-skill Legacy Skill routes rather than Native Plugin support.
• The README states that Legacy is a separate file-based implementation with a different supported data model. It explicitly says not to claim SQLite, five-tool MCP, Panel, or Plugin lifecycle support.
• The manual compatibility promise currently covers only local files and pasted text. PDF, email, provider exports, and hosted connectors are outside the supported scope.
• Enabling old collectors while the Plugin uses the same ~/.distilly/ namespace may write credential configuration and cross the Preview’s reviewed security boundary. The README also requires exactly one active distilly in each host discovery scope.

Legacy is therefore suitable for maintaining basic Skill discoverability, but not for replacing a verified Plugin or expanding the supported feature claims.

Yes, because Codex is currently the most completely verified host and supports the full workflow.

• The README section “The first usable flow” states that Codex verifies the complete path: importing materials, resolving a person, creating an evidence-bound briefing, committing a versioned Person Profile, submitting corrections, reviewing or rolling back candidates, and installing an approved Skill.
• Supported inputs are explicitly supplied TXT, Markdown, JSON, SRT/VTT files, pasted text, and public URLs. The system rejects symlinked selected files and duplicate file names.
• The model-facing surface is limited to five MCP tools: distilly_get, distilly_ingest, distilly_pending, distilly_commit, and distilly_correct. If a profile is too large, Distilly reports a capacity error instead of silently truncating it.

The README does not provide a complete Codex installation command, and the project has only one published release, v0.01. Therefore, this suitability applies mainly to the Developer Preview and controlled use, not to a guaranteed production-stable deployment.

It depends: OpenClaw is suitable for validating briefing transport and five-tool discovery, but not yet for treating the full long-term lifecycle as production-verified.

• The README records a 65,536-byte net transport budget for OpenClaw 2026.3.24 and says that a transport-capacity fixture exists for that version.
• OpenClaw loads the Claude-compatible bundle from ~/.openclaw/extensions/distilly; discovery can be checked with openclaw plugins inspect distilly --json.
• However, “The first usable flow” explicitly says that packaged restart, long-lived Skill, and uninstall lifecycle checks for OpenClaw remain separate and unverified.
• An unrecorded host version, release digest, tool descriptor, or serializer tuple returns host_unsupported before writing. Distilly does not automatically switch to the Legacy implementation.

Thus, the integration is justified for validating material ingestion and briefing transport. The current evidence is insufficient if the deployment depends on durable Skills, upgrades, or uninstall behavior.

It depends: Hermes has a documented installation path and a verified briefing-transport fixture, but its complete lifecycle is not verified and should not be treated as production-complete.

• The README specifies ~/.hermes/skills/distilly, ~/.distilly/bin/distilly-hermes, and ~/.hermes/config.yaml. It also disables resources and prompts so that exactly five tools remain exposed.
• Hermes v0.9.0 has a recorded net budget of 49,752 bytes, and the README says its briefing transport path is verified against that fixture.
• MCP discovery can be checked with hermes mcp test distilly, but the Host status table still marks restart, long-lived Skill, and uninstall lifecycle checks as pending.
• Unrecorded versions or serializer combinations are rejected before writing, and a failed Plugin preflight does not automatically switch to Legacy mode.

This makes Hermes suitable for controlled MCP and briefing integration, but not for assuming that persistence, upgrades, and uninstall behavior are already delivered capabilities.

It depends: Distilly’s local-first, versioned, and correction-aware model matches the requirement, but cross-host deployment is constrained by recorded fixtures, transport capacity, and incomplete lifecycle validation.

• The project insight describes local data as authoritative and supports evidence-bound briefings, versioned Person Profiles, explicit corrections, and local-panel approval, rejection, or rollback of candidate versions.
• Codex has the complete flow verified. OpenClaw 2026.3.24 and Hermes v0.9.0 have recorded briefing-transport fixtures, but long-lived Skills, complete restart behavior, and uninstall lifecycles are not all verified.
• The hosts have different transport budgets: 65,536 bytes for OpenClaw and 49,752 bytes for Hermes. When the complete value exceeds a verified budget, Distilly rejects the transfer and leaves stored data unchanged.
• The README requires keeping exactly one discoverable distilly instance and checking which copy was actually loaded. Local-first does not remove the user’s responsibility for permissions around local files, host processes, and public URLs.

It is therefore suitable for controlled cross-host profile reuse, but not for promising identical persistence, upgrade, and uninstall behavior across all hosts.