Repository navigation
TDD 0065: parent-session light-tier check (FR-86) - #184
Merged
cahenesy merged 4 commits intoOct 5, 2026
Merged
Conversation
Add tests/parent-session-check.test.sh: extracts the first fenced bash block after `<!-- tl:fr86-check -->` from /prd-author, /tdd-author and /build-tdds and runs it as the harness does (env -i, `bash <block>`, no args, nothing pre-sourced) against fixture transcripts and a Grok summary.json. Pins observations 1-8 of the TDD 0065 Verification plan: byte-identical blocks, silent above the light tier, the exact light and unreadable warning lines, fail closed on an unusable plugin root, Continue/Stop/fail-closed prose after the block, and no vendor-page fetch text. A missing marker, block or skill file is an infra FAIL. Register it in the implement-gate aggregator; a registered eval that is missing now fails instead of being skipped (L-011). Red: no skill carries the marker and models.sh has no tl_fr86_message. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Ki3TDnP2bNqqpEkFoZaEXq
Add tl_fr86_message to scripts/lib/models.sh (FR-86). It calls tl_parent_model once and reuses tl_model_tier: a readable parent above the light tier prints nothing; a light-tier parent prints one line naming the id; an unreadable parent prints one line carrying the reason from tl_parent_model's stderr (`unknown` when none). rc 0 in all three cases, rc 2 only when the library itself is unusable. No network and no capability comparison (ADR 0015). /prd-author (after step 0), /tdd-author (after the resume check) and /build-tdds (new step 1a, before the lock) carry the same `<!-- tl:fr86-check -->` block: it sources plugin-root.sh and models.sh itself, fails closed if either cannot be sourced, and calls tl_fr86_message in that same shell, so it works when the harness runs it as a fresh shell. The prose after it asks Continue / Stop on a printed line and stops on a non-zero exit. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Ki3TDnP2bNqqpEkFoZaEXq
Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Ki3TDnP2bNqqpEkFoZaEXq
Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Ki3TDnP2bNqqpEkFoZaEXq
cahenesy
deleted the
build/20261005-174458/0065-parent-session-model-check
branch
October 5, 2026 19:01
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.
Implements TDD 0065 (PRD f6ef178, ADR 0015). Second in the stack; 0064 is already merged, so this targets
master.What changed
New
tl_fr86_messageinscripts/lib/models.shreturns one of three results:There is no web fetch and no "most capable" comparison.
New check in
/prd-author,/tdd-authorand/build-tdds. Each skill gets the same<!-- tl:fr86-check -->block, byte for byte, with Continue/Stop prose and fail-closed sourcing. In/build-tddsit is a new## 1asection, before the lock.New
tests/parent-session-check.test.sh(68 checks). It extracts each skill's block and runs it in a cleanenv -ishell, the way the harness does. A test that only greps the text missed exactly this in the earlier 0064 build. In a scratch copy with themodels.shsource line removed from one block, 11 checks failed.Test-gate change.
tests/implement-gate.test.shnow fails when a registered eval is missing, where it used to skip it silently (L-011).Version 3.46.0 → 3.47.0, with a matching
releases.jsonentry.Gates
b3f5526 test(failing)→c7ecdfc feat; red on master: 27 fail)/build-tddsblock on this session printed nothingFollow-ups from review (not blocking)
[A-Za-z0-9._:@/+-]and cap its length./prd-authorand/tdd-authorthe check comes after step 0's Resume/Discard question, so Discard followed by Stop has already removed the draft. The TDD put it there; a revision should move the block before step 0./scripts/.... It fails closed, but the error message isn't clean.Merge note
Merge with a regular merge, not squash, so 0066 stacks cleanly. Don't run
claude plugin updateuntil 0066 is merged.🤖 Generated with Claude Code
https://claude.ai/code/session_01Ki3TDnP2bNqqpEkFoZaEXq