You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Make review-path lists compact and human-readable #35
Please revive #15’s compact, human-readable YAML path tree; #14’s closure covers relay recovery, not this format.
0.9.4 and current main still use arrays; preserve exact coverage and coordinate reader compatibility.
Illustrative repeated-prefix worst case (review.paths excerpt, not executed output; have wraps the current example):
Thanks for following up on this, and for the work you put into #15.
Help me understand the need here. What's the situation where you or your agents are reading these paths? For example, are you debugging queued moves, reviewing what a verification covered, or seeing agents spend tokens on them? Also curious how your reviews reach 1,000+ paths. With that context, we may find a better solution.
Are your agents running pullboard show <id> --json? If that's the token cost, the better fix may be adding a variant option, like a compact view, without changing the relay format. YAML can be delicate and easily break, that's my main concern.
E.g., something like this (doesn't exist yet):
pullboard show <id> --compact
For a tighter feedback loop, join our Discord. The link is on the website.
Thanks—the concrete use case was diagnosing a queued verify move and inspecting its exact reviewed-file coverage.
The 1,420 paths matched the cumulative item diff: 1,394 were tracked evidence/capture files (results, stdout/stderr, etc.), and 26 were source/test/template/document files. That bulk evidence was our workflow problem, not 1,420 handwritten changes or a typical review size.
Two concerns got conflated: #14 measured a transport problem (133,993 bytes for the path array; 136,018 for the plaintext move). #35 is about readable, less repetitive output for people and agents. We haven't measured model-token savings or established that show <id> --json is the token bottleneck; the examples above are illustrative, not a token benchmark.
Your proposed show <id> --compact sounds like a good fit for the presentation goal, with exact paths still available for drill-down. YAML isn't a requirement, and this issue needn't require a relay-format change. Reducing generated evidence in our submissions is a separate part of the answer.
Activity
inverted-capital commentedon Nov 21, 2026
It just seemed easier to read and less tokens / lower message size 🤷
coreyolson commentedon Nov 22, 2026
Thanks for following up on this, and for the work you put into #15.
Help me understand the need here. What's the situation where you or your agents are reading these paths? For example, are you debugging queued moves, reviewing what a verification covered, or seeing agents spend tokens on them? Also curious how your reviews reach 1,000+ paths. With that context, we may find a better solution.
Are your agents running
pullboard show <id> --json? If that's the token cost, the better fix may be adding a variant option, like a compact view, without changing the relay format. YAML can be delicate and easily break, that's my main concern.E.g., something like this (doesn't exist yet):
For a tighter feedback loop, join our Discord. The link is on the website.
McHersheys commentedon Nov 22, 2026
Thanks—the concrete use case was diagnosing a queued
verifymove and inspecting its exact reviewed-file coverage.The 1,420 paths matched the cumulative item diff: 1,394 were tracked evidence/capture files (results, stdout/stderr, etc.), and 26 were source/test/template/document files. That bulk evidence was our workflow problem, not 1,420 handwritten changes or a typical review size.
Two concerns got conflated: #14 measured a transport problem (133,993 bytes for the path array; 136,018 for the plaintext move). #35 is about readable, less repetitive output for people and agents. We haven't measured model-token savings or established that
show <id> --jsonis the token bottleneck; the examples above are illustrative, not a token benchmark.Your proposed
show <id> --compactsounds like a good fit for the presentation goal, with exact paths still available for drill-down. YAML isn't a requirement, and this issue needn't require a relay-format change. Reducing generated evidence in our submissions is a separate part of the answer.