Support batch transcript fetching from piped stdin#180
Merged
Conversation
`transcripts get` now fetches multiple transcripts when ids are piped on
stdin and the positional id is omitted, so the collection commands compose
as a Unix pipeline:
assembly transcripts list --json | assembly transcripts get -o text
A new `client.parse_transcript_ids` accepts the shapes a pipeline naturally
produces — the JSON array from `transcripts list --json`, a single transcript
JSON object, or plain text with one id per line (`jq -r '.[].id'`) — so the
pipeline works with or without jq. In stdin (batch) mode `--json` emits one
NDJSON `{"type": "transcript", ...}` record per id, mirroring `transcribe`
batch, while a single positional fetch keeps its existing output shape.
https://claude.ai/code/session_01GdpgpKDUJCNvECP5ttowst
Mirrors the map-reduce LLM vocabulary `transcribe` gained in #179, completing the pipeline `transcripts list | transcripts get` so fetched transcripts can be summarized or aggregated without a second tool: # map: summarize each transcript in a piped list assembly transcripts list --json | assembly transcripts get --llm "Summarize this call" # reduce: one ranking across all of them assembly transcripts list --json | assembly transcripts get --llm-reduce "Rank these worst-to-best" `--llm` runs a per-transcript chain (server-injected by id via `llm.run_chain_steps`); `--llm-reduce` runs one chain over all fetched transcripts (`llm.run_chain`), emitting the same additive `{"type":"reduce",...}` NDJSON record transcribe does. A single positional id folds the reduce prompts into the `--llm` chain (nothing to aggregate), matching transcribe's single-source behavior. Human reduce keeps stdout clean by suppressing the per-transcript output; --json keeps the per-id stream and appends the reduce record. Reuses core/llm.py and transcribe's render_transform_steps — no new engine. The get/list pipeline tests move to tests/test_transcripts_pipeline.py to keep both test modules under the 500-line gate. https://claude.ai/code/session_01GdpgpKDUJCNvECP5ttowst
![aikido-pr-checks[bot]](https://avatars.githubusercontent.com/in/898896?s=60&v=4){kind=small}
Comment on lines
+152
to
+158
| if output_field is not None: | ||
| # -o wins over the chain, matching `transcribe` deliver_result precedence. | ||
| output.emit_text( | ||
| client.select_transcript_field( | ||
| transcript, output_field, chars_per_caption=chars_per_caption | ||
| ) | ||
| ) |
There was a problem hiding this comment.
_deliver_transcript ignores suppress when output_field is set, so per-transcript output still prints during reduce-suppressed runs, contradicting the function’s suppression logic.
Show fix
Suggested change
| if output_field is not None: | |
| # -o wins over the chain, matching `transcribe` deliver_result precedence. | |
| output.emit_text( | |
| client.select_transcript_field( | |
| transcript, output_field, chars_per_caption=chars_per_caption | |
| ) | |
| ) | |
| if output_field is not None: | |
| if not suppress: | |
| output.emit_text( | |
| client.select_transcript_field( | |
| transcript, output_field, chars_per_caption=chars_per_caption | |
| ) | |
| ) |
Details
✨ AI Reasoning
The function is designed to optionally skip per-transcript emission when a reduce step will follow. That behavior is implemented for the transform and default transcript branches, but not for the single-field branch. As a result, when reduce mode intends to suppress per-item output, selecting a field still prints each item, violating the function’s own contract and making the suppression path inconsistent.
Reply @AikidoSec feedback: [FEEDBACK] to get better review comments in the future.
Reply @AikidoSec ignore: [REASON] to ignore this issue.
More info
`_deliver_transcript`'s `-o` branch ignored the `suppress` flag, so `transcripts get -o text --llm-reduce …` in human mode printed each transcript's field *and* the reduce result. Guard the field emit with `if not suppress:` so it matches the transform/default branches: under a pending human reduce only the aggregate reaches stdout, while each field still feeds the reduce. (Flagged in review.) https://claude.ai/code/session_01GdpgpKDUJCNvECP5ttowst
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Enable the
transcripts getcommand to read multiple transcript IDs from piped stdin, composing naturally withtranscripts list --jsonand other upstream commands.Summary
This change makes
transcripts getaccept zero or more transcript IDs from stdin when no positional argument is provided, enabling pipelines like:assembly transcripts list --json | assembly transcripts get -o textThe command maintains backward compatibility: a positional ID still works as before, and the output shape differs between single-fetch (unchanged) and batch modes (NDJSON with
--json, plain text otherwise).Key Changes
_resolve_ids()helper: Determines whether to fetch a single positional ID or read multiple IDs from stdin. RaisesUsageErrorif neither is provided or stdin contains no valid IDs, preventing hangs on interactive invocation._emit_transcript()helper: Centralizes transcript output logic, handling four distinct modes:-o field: raw field output (pipeline-friendly)--json: full SDK payload (matchestranscribe --json)--json: NDJSON records tagged with"type": "transcript"(CLI-wide discriminator)--json: plain text (one transcript per line)parse_transcript_ids()inclient.py: Flexible stdin parser accepting:transcripts list --json(extractsidfield)idfield)jq -r '.[].id')Argument change:
transcript_idis now optional (str | None), with help text clarifying stdin fallback.Test coverage: Added comprehensive tests for ID parsing, batch fetching, error cases (no ID + no stdin, empty stdin), and output format validation (single-fetch JSON excludes
"type"wrapper; batch JSON includes it).Implementation Details
batchflag returned by_resolve_ids()controls output formatting, ensuring NDJSON wrapping only appears in true batch mode.https://claude.ai/code/session_01GdpgpKDUJCNvECP5ttowst