feat(channels): add Fish Audio voice/telephony channel ingest (#5045) - #5055
Conversation
Visual diffComparing 46 of 70 comparison(s) flagged (>1% pixel diff).
Folder: 90a4fffb3271. Full PNGs also attached as a workflow artefact. Generated by visual-diff bot. Pixel diffs >1% flagged; eyeball the table before merging. This check is non-blocking — fail = bot bug, not a code problem. |
|
Autonomous maintainer pass (2026-08-22): converted this from draft to ready-for-review. CI is 100% green (35 checks). Branch rebased onto latest main (which just received the i18n doc sync merge -- no conflicts, clean state). One human action still required before merge: confirm the directory name Once the directory name is confirmed, this is ready to squash-merge — CI is green, code is correct, closes #5045. Generated by Claude Code |
✅ Drift Bot (ClawMetry): no drift detectedDrift Bot analyzed the changed files against this project's blueprints and requirements and found no drift. |
1 similar comment
✅ Drift Bot (ClawMetry): no drift detectedDrift Bot analyzed the changed files against this project's blueprints and requirements and found no drift. |
|
PR janitor check (2026-08-22): CI is showing
Action needed: please re-run the failed job (Actions → "Re-run failed jobs") to unblock CI. This PR's code is unrelated to the timeout. Generated by Claude Code |
|
✨ auto-fixed: raised Generated by Claude Code |
|
![8090-software-factory[bot]](https://avatars.githubusercontent.com/in/1847834?s=60&v=4){kind=small}
|
Autonomous maintainer pass (2026-08-22): status check. CI: 35/35 checks green. All functional tests, sync tests, E2E, entitlements, and install smoke tests pass. Visual diff flagged 36/66 comparisons, but the bot marks that check non-blocking ("fail = bot bug, not a code problem"). Drift bot flagged 3 doc-only findings (no blueprint for Fish Audio adapter) — those are informational, not blockers. Still blocking merge: directory name confirmation. The PR body (and prior bot comment) flag this clearly: the four What's needed: verify Generated by Claude Code |
|
✨ auto-fixed: branch was 2 commits behind main ( Generated by Claude Code |
|
|
Maintenance pass review — CI is fully green (35/35 checks) and the code is clean, but this PR explicitly asks for human verification before merging:
I cannot verify the directory name without access to a live OpenClaw instance or the private OpenClaw source. Once you confirm the path is correct, the PR is ready to squash-merge. No code changes needed from this pass. Generated by Claude Code |
|
Automated maintenance check — all 35 CI checks are green on this PR. Blocked on human verification: The PR body correctly flags this: the directory name Once the directory name is confirmed (or corrected), this is safe to merge — CI is clean across all 35 jobs. Generated by Claude Code |
|
CI is fully green (all 35 checks pass). The implementation is solid -- four-point adapter checklist complete in 18 lines. Blocking item for merge: The PR body itself flags this (reviewer action required): the string Once the directory name is confirmed correct, this is ready to squash-merge. Generated by Claude Code |
|
|
|
This PR is blocked by the Drift Bot check, which is an external product-record gate run by the 8090 Software Factory. The check fails when a PR touches product code but no linked Factory record (PRD/blueprint) has been approved, or when the body contains no This cannot be unblocked by code changes — the product review team needs to approve the associated Factory record, or the PR description needs a Action needed by the author: either approve the Factory record in 8090 Software Factory, or add Generated by Claude Code |
|
PR sweep note (automated): This PR is blocked for two reasons:
No code changes applied. Needs human sign-off on the Generated by Claude Code |
|
blocked on author decision — skipping (auto-mergeability sweep) The PR body explicitly notes "Reviewer action required: verify the directory name" — specifically whether Generated by Claude Code |
|
PR sweep bot: Re-triggered the failing E2E Gate run (run 34494594947) via Root cause (confirmed from CI logs): The E2E Gate fails because The blocker is entirely at factory.8090.ai: one or more blueprints reference code paths that don't match what this PR introduced. Update the relevant blueprint(s) at factory.8090.ai to match the new code, then Drift Bot will clear and the E2E Gate will pass. If the re-triggered run still fails, the fix must happen at factory.8090.ai before this PR can merge. Generated by Claude Code |
|
| @@ -6487,6 +6487,7 @@ def sync_openclaw_claude_sessions_via_index( | |||
| "nextcloudtalk", | |||
There was a problem hiding this comment.
The blueprint does not document the _CHANNEL_DIRS communication channel adapter ingest mechanism. The codebase implements 24 channel adapters where the sync daemon walks ~/.openclaw//*.jsonl directories to ingest transcripts, but this core architectural pattern is absent from the blueprint specification.
| if fast is not None: | ||
| return jsonify(fast) | ||
| return _d._generic_channel_data("buzz") | ||
|
|
There was a problem hiding this comment.
The blueprint does not document the HTTP endpoint contract for communication channel adapters. The codebase implements /api/channel/* endpoints for 24 adapters following a DuckDB fast-path plus legacy fallback pattern, but this architectural contract is undocumented.
| # We kept it as a list (rather than dropping to a single subscribe call) | ||
| # so a future per-channel subscribe API can plug in without touching | ||
| # the call sites in sync.py. | ||
| # |
There was a problem hiding this comment.
The blueprint does not document the gateway WebSocket tap architecture that enables live channel event ingest. The codebase implements CHANNEL_NAMES subscriptions for 24 adapters, but this core adapter bootstrap pattern is absent from the blueprint specification.
|
| @@ -6487,6 +6487,7 @@ def sync_openclaw_claude_sessions_via_index( | |||
| "nextcloudtalk", | |||
There was a problem hiding this comment.
The blueprint does not document the _CHANNEL_DIRS communication channel adapter ingest mechanism. The codebase implements 24 channel adapters (including Fish Audio added in this PR) where the sync daemon walks ~/.openclaw//*.jsonl directories to ingest transcripts, but this core architectural pattern is absent from the blueprint specification.
| if fast is not None: | ||
| return jsonify(fast) | ||
| return _d._generic_channel_data("buzz") | ||
|
|
There was a problem hiding this comment.
The blueprint does not document the HTTP endpoint contract for communication channel adapters. The codebase implements /api/channel/* endpoints for 24 adapters (including Fish Audio added in this PR) following a DuckDB fast-path plus legacy fallback pattern, but this architectural contract is undocumented.
| # We kept it as a list (rather than dropping to a single subscribe call) | ||
| # so a future per-channel subscribe API can plug in without touching | ||
| # the call sites in sync.py. | ||
| # |
There was a problem hiding this comment.
The blueprint does not document the gateway WebSocket tap architecture that enables live channel event ingest. The codebase implements CHANNEL_NAMES subscriptions for 24 adapters (including Fish Audio added in this PR), but this core adapter bootstrap pattern is absent from the blueprint specification.
|
Sweep bot status — Drift Bot is blocking this PR (operator action required) Drift Bot is reporting 5 drift findings on the current head ( Root cause: The blueprint(s) referenced by this PR haven't been updated to describe the new/changed code. Drift Bot validates every diff against the stored blueprint spec, so it will keep firing until the blueprints are updated. What's needed (operator action only — cannot be done from the repo):
Once the blueprint(s) match the implementation, Drift Bot will pass and the E2E Gate will unblock. Generated by Claude Code |
|
| @@ -6487,6 +6487,7 @@ def sync_openclaw_claude_sessions_via_index( | |||
| "nextcloudtalk", | |||
| "clickclack", | |||
There was a problem hiding this comment.
The blueprint does not document the _CHANNEL_DIRS mechanism that the sync daemon uses to ingest transcripts from communication channel adapters. The codebase implements 24 channel adapters (including Fish Audio added in this PR) where the daemon walks ~/.openclaw//*.jsonl directories, but this core ingest pattern is undocumented in the blueprint.
| return jsonify(fast) | ||
| return _d._generic_channel_data("buzz") | ||
|
|
||
|
|
There was a problem hiding this comment.
The blueprint does not document the HTTP endpoint contract for communication channel adapters. The codebase implements /api/channel/* endpoints for 24 adapters (including Fish Audio added in this PR) following a DuckDB fast-path plus legacy fallback pattern, but this architectural contract is undocumented.
| @@ -106,6 +111,8 @@ | |||
| "synologychat", | |||
| "nextcloudtalk", | |||
| "clickclack", | |||
There was a problem hiding this comment.
The blueprint does not document the gateway WebSocket tap architecture that enables live channel event ingest. The codebase implements CHANNEL_NAMES subscriptions for 24 adapters (including Fish Audio added in this PR), but this core adapter bootstrap pattern is absent from the blueprint specification.
…isled Drift Bot found five, all the same omission: adding the 24th channel exposed that the whole subsystem has no product record -- _CHANNEL_DIRS, the /api/channel/* contract, the gateway WebSocket subscription, and the entitlement model. Documenting only Fish Audio would leave the identical finding waiting for the 25th. The Runtime and Session Observability blueprint now records it as six contracts. The load-bearing one: a channel is FOUR independent lists, and one in three of them is broken in a way that looks like a different bug each time. sync._CHANNEL_DIRS historical transcripts ingest gateway_tap.CHANNEL_NAMES live events arrive entitlements.ALL_CHANNELS the UI and entitlement can see it routes/channels.py the endpoint exists Miss the gateway list and live messages never arrive while history works fine. Miss the catalogue and the channel is invisible even with rows in the store. Neither name list may be derived from the other by string munging, because the disk name and the wire name legitimately differ -- fish-audio on disk, fishaudio on the wire, in this very PR. The comment above _CHANNEL_DIRS said adding a directory needed "no further wiring required". That is false in exactly the way above, and it is the sentence that would make the next person ship a half-wired channel. It also still said "21 adapter routes" when there are 24. Both corrected, and the comment now names all four lists with the symptom for each. Verified Fish Audio is in all four (it is), the catalogue guard passes (27 tests), and the count guard reads 24 across every surface. An ADR records why these stay four explicit lists rather than one derived registry, and states plainly that an independent omission is still caught by review rather than by a check -- an accepted gap, written down instead of implied. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Xb6A5G74JiMe3zHFs1JZEP
|
| # We kept it as a list (rather than dropping to a single subscribe call) | ||
| # so a future per-channel subscribe API can plug in without touching | ||
| # the call sites in sync.py. | ||
| # |
There was a problem hiding this comment.
The blueprint specifies that a channel is only shipped when it appears in all four lists including gateway_tap.CHANNEL_NAMES for live WebSocket event subscription. The buzz channel is missing from CHANNEL_NAMES, meaning live messages would never arrive even though the channel has an endpoint in routes/channels.py and exists in the other lists.
There was a problem hiding this comment.
This PR already fixes the gap. The diff adds both "buzz" and "fishaudio" to CHANNEL_NAMES together — see clawmetry/gateway_tap.py line 114 in the current head (72eecaf). On main, "buzz" has an endpoint in routes/channels.py and entries in _CHANNEL_DIRS and ALL_CHANNELS, but was indeed missing from CHANNEL_NAMES (a pre-existing half-wiring). This PR ships both channels fully wired across all four lists.
Generated by Claude Code
Drift Bot reported that `buzz` is missing from gateway_tap.CHANNEL_NAMES, so its live messages never arrive. That finding is wrong: buzz is at gateway_tap.py:114, and all three lists hold 24 entries with no gaps in either direction. I did not change code to satisfy it. But it points at a real weakness I wrote into my own ADR one commit ago: the four-list rule was enforced by review rather than by a check, which is exactly the kind of claim that cannot be trusted. It is a check now. tests/test_channel_four_lists.py asserts every catalogued channel appears in _CHANNEL_DIRS and CHANNEL_NAMES, that neither ingest list carries an orphan the catalogue does not know about, that every channel has a CHANNEL_LABELS entry, and that the three lists are the same size. Comparison is on a normalised form, because the disk name and the wire name legitimately differ (fish-audio vs fishaudio) and deriving one list from the other is what the blueprint forbids. Proven RED against the exact condition Drift Bot described: deleting "buzz" from CHANNEL_NAMES reproduces its message verbatim -- in ALL_CHANNELS but not in gateway_tap.CHANNEL_NAMES: ['buzz']. Live messages for these never arrive, while historical transcripts ingest normally -- which reads as an ingest bug, not a missing subscription. -- and passes on the real tree, which is the disproof. The tuples are text-parsed rather than imported, so this runs in the plain lint job: sync.py and gateway_tap.py pull in the daemon's dependencies, and a guard that needs them would have to live in a heavier job or silently skip. Named in ci.yml, in a job that can actually run it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Xb6A5G74JiMe3zHFs1JZEP
✅ Drift Bot (ClawMetry): no drift detectedDrift Bot analyzed the changed files against this project's blueprints and requirements and found no drift. |
|
Live in PyPI The first Verified from the wheel's own bytesThe corrected A correction to this PR's own descriptionThe CHANGELOG and PR text said transcripts ingest from The example came from a pre-existing comment in
No cloud promotion needed: channel ingest is node-side and reaches machines through the daemon's own PyPI self-update. |
1 participant
No-PRD: Mechanical wire-up following the established channel adapter pattern (same as ClickClack, Buzz, etc.); no new product decisions -- Fish Audio channel support was planned in #5045.
Summary
Fish Audio's voice/telephony channel writes session transcripts to
~/.openclaw/fishaudio/butsync.py's_CHANNEL_DIRShad no entry for it, so all transcripts were silently skipped. This wires up the full four-point adapter checklist -- ingest, catalog, UI label, and HTTP endpoint -- in 18 lines across 4 files.Changes
clawmetry/sync.py: add"fishaudio"to_CHANNEL_DIRSso the sync daemon walks~/.openclaw/fishaudio/*.jsonlon each cycleclawmetry/entitlements.py: add"fishaudio"toALL_CHANNELS(satisfies the pin test intests/test_entitlement_channel_catalog.py) and"Fish Audio"toCHANNEL_LABELSclawmetry/static/js/app.js: add icon (🎤) and brand color (#FF5A36) to_channelIcons/_channelColorsroutes/channels.py: add/api/channel/fishaudioendpoint following the identical DuckDB fast-path + legacy fallback pattern asapi_channel_clickclack()/api_channel_buzz()Test plan
python3 -m pytest tests/test_entitlement_channel_catalog.py -q--test_all_channels_matches_sync_channel_dirsmust pass (verified manually: both tuples now have 24 matching entries).jsonlfile under~/.openclaw/fishaudio/and confirmclawmetry syncingests it;GET /api/channel/fishaudioshould return rowsfishaudiomatches what OpenClaw actually writes to disk -- if OpenClaw usesfish_audio,talk, or another path, update the four"fishaudio"string literals accordingly before mergingBot meta
Draft PR opened autonomously based on the plan in #5045. Marked draft for human review -- mark Ready for Review once the directory name is confirmed.
Closes #5045