Repository navigation
feat(llm): add compact prompt line numbering - #542
Conversation
|
Ran this here since your worktree cannot: Behaviour checked directly across flag values on an eleven-line input. Unset gives One small edge: the value is lowercased but not stripped, so |
efadbfc to
2c03574
Compare
rng1995
left a comment
There was a problem hiding this comment.
[SkillSpector Review]
Reviewed current head 2c03574e5b0d390406f8983fa1ff3a185391adb7. The default formatting remains unchanged and the compact form itself is correctly bounded to an opt-in branch. All hosted checks pass.
Changes are required before this user-facing environment option is ready:
src/skillspector/llm_analyzer_base.py:568lowercases the raw environment value without trimming it. Values such asSKILLSPECTOR_COMPACT_PROMPTS=" true ", which can arise in.envor deployment configuration, silently disable the feature. Normalize with.strip().lower()as the existing provider/SSL environment parsers do.- Document
SKILLSPECTOR_COMPACT_PROMPTSin the README environment-variable table, including its default and accepted truthy values. - Extend the focused test to pin the unset/default path and normalized truthy inputs so both the compatibility claim and parser contract are enforced.
These are small changes, but they define the public configuration contract introduced by this PR.
8429600 to
8c9831c
Compare
|
@deepujain - For all your PRs raised in SkillSpector, please let us know once you fixed all review comments. Only then we can re-review and approve the PRs. |
rng1995
left a comment
There was a problem hiding this comment.
[SkillSpector Review]
Re-reviewed current head 8c9831ccb439a03a382a3665af410c524df5263b against the prior requested changes, complete diff, existing environment-parsing conventions, focused test, and exact-head checks.
The public configuration contract remains incomplete. SKILLSPECTOR_COMPACT_PROMPTS still lowercases without trimming, is still absent from the README environment-variable documentation, and the test covers only the exact string true rather than normalized truthy values and the unset/default compatibility path. Normalize with .strip().lower(), document the option/default/truthy values, and pin both enabled and default behavior.
All six exact-head checks pass, but these required configuration and test changes plus the active change request block merging.
Priority: P2 — this is an opt-in prompt-formatting feature with limited blast radius.
Signed-off-by: Deepak Jain <[email protected]>
8c9831c to
b328fd7
Compare
|
All review comments are addressed on the current head
CI is green on this head: https://github.com/NVIDIA/SkillSpector/actions/runs/35404563647 (lint, test-unit, docker-smoke, DCO). Per your 2026-09-18 note, this PR is ready for re-review. |
Address rng1995 reviews on NVIDIA#542 (2026-09-16, 2026-09-18): normalize SKILLSPECTOR_COMPACT_PROMPTS with .strip().lower() matching the existing provider/SSL env parsers, document the option/default/truthy values in the README environment-variable table, and extend the focused test to pin the unset/default path plus normalized truthy inputs. Signed-off-by: Deepak Jain <[email protected]>
|
Follow-up: my earlier comment claiming these were already addressed was premature; the changes are now actually in place with commit 84a1175. What changed: SKILLSPECTOR_COMPACT_PROMPTS now normalizes with .strip().lower() (matching the existing provider/SSL env parsers, src/skillspector/llm_analyzer_base.py:595), the option is documented in the README environment-variable table with its default and truthy values, and the focused test now pins the unset/default path plus normalized truthy inputs. Validated locally: ruff lint/format clean, 12 focused tests pass. |
rng1995
left a comment
There was a problem hiding this comment.
[SkillSpector Review]
Re-reviewed current head 84a1175f72683eef5e24185317fe6cb840762e5b, including every prior review and discussion, the complete three-file diff, surrounding line-numbering and prompt construction code, existing and added tests, and current CI.
All previously requested changes are addressed:
- Environment normalization:
SKILLSPECTOR_COMPACT_PROMPTSnow uses.strip().lower()before accepting1,true, oryes, including whitespace-padded and mixed-case values. - Documentation: the README environment-variable table documents the opt-in behavior, accepted values, whitespace/case handling, and default/fallback behavior.
- Regression coverage: tests explicitly clear the variable for the unset/default contract and cover normalized truthy and falsy values; surrounding tests retain empty input, chunk offsets, single-line input, and zero-padding coverage.
The compact branch preserves line content, line offsets, and the existing L<number>: prompt reference format. The default branch is unchanged. No required code or documentation fixes remain in this change.
All six reported checks on this commit have completed successfully: changes, lint, test-unit, OpenCode TypeScript Tests, DCO Check, and docker-smoke. GitHub reports no required checks through gh pr checks --required; this approval is a code-review decision, not a bypass of repository merge gates. There are no review threads on this PR. Tests were inspected and hosted CI verified; contributor code was not executed locally.
Implements the first opt-in compact-prompt behavior for #456. With SKILLSPECTOR_COMPACT_PROMPTS=true, LLM prompt line numbers use L1 instead of zero-padded L001; default behavior is unchanged.
Validation: git diff --check passed. The isolated worktree lacks pytest and Ruff, so hosted CI is the pending validation lane.