fix(agent): keep Life CLI results out of long OMP tool previews #118

New issue

Closed

opened 2026-09-23 23:45:28 +00:00 by yuris · 0 comments

yuris commented

2026-09-23 23:45:28 +00:00

Owner

Why

A raw life agenda or tasks list JSON envelope contains full item details. OMP displays the tool output and spills long results to artifact links; reading those links creates another visible tool card. The user wants answers, not a stream of private calendar details and large command outputs. A skill instruction alone cannot promise to hide OMP's own tool cards. We need a verified invocation pattern that reduces what reaches the preview, not a claim that the CLI stops returning data.

Scope

Determine what OMP currently displays for Life CLI calls and whether it offers a per-tool or per-skill preview suppression setting. Do not invent an OMP configuration key. If no supported setting exists, explicitly document that complete suppression is unavailable and use the smallest reliable output path.
Capture full JSON internally, parse status, schema_version, data, warnings, and error; emit only the fields needed for the user's question and bounded item rows. Prefer existing jq or runtime facilities over a new wrapper or dependency. Do not leak raw envelope, credentials, sensitive item descriptions, or artifact links as a default.
Preserve the Life process exit code before parsing. A pipeline that reports jq success after life fails is not acceptable. Keep warnings, incomplete coverage, request/operation IDs needed for recovery, and partial/unknown results visible in the compact summary; never turn a failed or partial write into a success.
Apply the pattern to agenda, tasks list, item get, next, and write result handling in the agent's operating guidance. Long results must be summarized with honest truncation and a way to request more detail deliberately.

Acceptance

Demonstrate OMP previews for a long agenda and task list are bounded, with no automatic artifact read, while the final answer still gives accurate as-of/coverage and relevant items.
Demonstrate a CLI nonzero exit and partial/unknown envelope preserve their status and recovery identifiers; no silent jq-success masking or discarded warning.
Record the observed OMP limitation if a tool card itself cannot be hidden. This issue is agent presentation only; do not shrink the Life API JSON contract or daemon payloads.

## Why A raw life agenda or tasks list JSON envelope contains full item details. OMP displays the tool output and spills long results to artifact links; reading those links creates another visible tool card. The user wants answers, not a stream of private calendar details and large command outputs. A skill instruction alone cannot promise to hide OMP's own tool cards. We need a verified invocation pattern that reduces what reaches the preview, not a claim that the CLI stops returning data. ## Scope - Determine what OMP currently displays for Life CLI calls and whether it offers a per-tool or per-skill preview suppression setting. Do not invent an OMP configuration key. If no supported setting exists, explicitly document that complete suppression is unavailable and use the smallest reliable output path. - Capture full JSON internally, parse status, schema_version, data, warnings, and error; emit only the fields needed for the user's question and bounded item rows. Prefer existing jq or runtime facilities over a new wrapper or dependency. Do not leak raw envelope, credentials, sensitive item descriptions, or artifact links as a default. - Preserve the Life process exit code before parsing. A pipeline that reports jq success after life fails is not acceptable. Keep warnings, incomplete coverage, request/operation IDs needed for recovery, and partial/unknown results visible in the compact summary; never turn a failed or partial write into a success. - Apply the pattern to agenda, tasks list, item get, next, and write result handling in the agent's operating guidance. Long results must be summarized with honest truncation and a way to request more detail deliberately. ## Acceptance - Demonstrate OMP previews for a long agenda and task list are bounded, with no automatic artifact read, while the final answer still gives accurate as-of/coverage and relevant items. - Demonstrate a CLI nonzero exit and partial/unknown envelope preserve their status and recovery identifiers; no silent jq-success masking or discarded warning. - Record the observed OMP limitation if a tool card itself cannot be hidden. This issue is agent presentation only; do not shrink the Life API JSON contract or daemon payloads.

yuris referenced this issue from a commit

2026-09-24 00:04:39 +00:00

feat(cli): add compact output mode for read commands

yuris referenced this issue from a pull request that will close it,

2026-09-24 00:04:45 +00:00

feat(cli): add compact output mode for read commands #120

yuris closed this issue

2026-09-24 07:22:05 +00:00