Repository navigation
feat: Port igrsoft v3.36.0 learnings - budgets, triage, evidence, lint - #10
Merged
Merged
Conversation
Sizing table verified already current (DR0 at every tier) - version refs only. PL0-stamps note now names metadata.test_mode and ui_visual_check (N/A for systems work, stays false). CLI evidence gains a freshness rule: cli-fallback transcripts produced this run only - igrsoft QA direct-reads and cross-checks against build-evidence log paths; stale or duplicated transcripts re-open DV. Historical CHANGELOG/MEMORY refs untouched.
The read->merge->temp->fsync->rename prose is gone. Primary: run igrsoft state-patch.sh --stage <CODE> --prev <PREV> when the dispatcher supplies its path (task.metadata.state_patch_script). Degrade: skip silently, no hand-rolled merge - Layers 2/3 repair from frontmatter, whose emission stays unconditional.
Benchmark-driven: AR-companion spawn tripled stage cost and DV ran 122K out-tokens/123 tool calls vs 77K/67 bare for equal results. Budgets cap artifact lines, return tokens, and tool calls (DV in _base, AR, DV-support, DR-support) with Build Evidence exempt from every cap; system-architector self-limits to Quick Recommendation Mode at Low complexity (0-10), mode triggers outranking an inferred low score.
Ported from igrsoft via apple-developer. section-lint: leaf sections <=1000 chars, warn-only baseline (pre-existing prose debt incl. references/; burn-down separate). desc-lint: two-tier caps - agents/ commands <=250, skills <=600 (regression brake above the 515 worst; tightening needs eval evidence). Wired into scripts/test.sh (desc fatal, section warn even under --strict); bats self-tests picked up by CI.
igrsoft v3.36.0 compatibility port. Version consistent across plugin.json, marketplace.json, README, MEMORY, CHANGELOG. section-lint baseline recorded: 491 sections over cap across 101 files (warn-only; burn-down tracked separately; scope includes references/). validate.sh: 0 errors, 4 pre-existing warnings.
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
1 participant
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.
Motivation
Port the company-workflow v3.36.0 session learnings (paired A/B benchmark + ov151 evidence-integrity findings) into system-developer, mirroring the apple-developer port (IGRSoft/apple-developer#119) adapted to systems work. Benchmark evidence: AR-companion spawn tripled stage cost ($5.36 vs $1.57 bare); DV ran 122K out-tokens/123 tool calls vs 77K/67 bare for equal results.
Changes
test_mode,ui_visual_check— N/A for CLI work, stays false).cli-fallbacktranscripts produced this run only, never reused; QA direct-reads and cross-checks against### build-evidencelog paths (the systems analog of the ov151 evidence-integrity gate). UI live-drive/hittability gates deliberately not ported — no hook in systems work.state-patch.sh --stage <CODE> --prev <PREV>when the dispatcher supplies its path, silent-skip degrade otherwise (Layers 2/3 repair from unconditional frontmatter)._base), AR, DV-support, DR-support blocks with Build Evidence exempt from every cap; system-architector self-limits to Quick Recommendation Mode at Low complexity (0-10).scripts/section-lint.sh(warn-only baseline: 491 sections over cap / 101 files, references/ in scope) +scripts/desc-lint.sh(two-tier 250/600 caps, green today) wired intoscripts/test.shwith bats self-tests; existing CI picks them up.validate.sh: 0 errors.Notes
Closes #9