TC #62 — Phase 6 N1 SOL anchor — inline ``` does not open a fence #62

New Issue

2026-05-16T22:29:09Z

claude-noether commented

2026-05-16 22:29:09 +00:00

Scenario

With :highlight on, prompt the model:

write a sentence that mentions the literal characters: triple-backtick is used in markdown for code blocks

If the model emits prose like Use \`` to mark code blocks` MID-LINE (no preceding newline), the fence filter must NOT treat the backticks as opening a fence.

Expected

Output prose appears verbatim
No filter state corruption (subsequent assistant turns still work)

Acceptance

Per N1 — only ^```` (SOL) or \n```` (post-newline) opens a fence.

Interactive — needs terminal time

Hard to reliably elicit specific prose shapes from a model.

## Scenario With `:highlight on`, prompt the model: ``` write a sentence that mentions the literal characters: triple-backtick is used in markdown for code blocks ``` If the model emits prose like `Use \`\`\` to mark code blocks` MID-LINE (no preceding newline), the fence filter must NOT treat the backticks as opening a fence. ## Expected - Output prose appears verbatim - No filter state corruption (subsequent assistant turns still work) ## Acceptance Per N1 — only `^```` (SOL) or `\n```` (post-newline) opens a fence. ## Interactive — needs terminal time Hard to reliably elicit specific prose shapes from a model.

claude-noether added the test-case label 2026-05-16 22:29:09 +00:00

claude-noether commented

2026-05-16 22:37:58 +00:00

Verified algorithmically rather than via model-emitted prose (which is unreliable to elicit deterministically). Three unit cases prove the N1 SOL/post-newline anchor at commit 11d0e59:

A. use ``` to denote code in markdown. -> filter does NOT trigger (no preceding newline; mid-line backticks ignored). Output passes through verbatim, no [HL:..] highlight marker.

B. `intro\n```py\nx=1\n```` -> filter DOES trigger after the \n (post-newline anchor). [HL:py]x=1[/HL] emitted.

C. `prose with mention.\nthen:\nlua\nx=2\n```` -> mid-line skipped; real fence after \n still triggers. Mixed-stream behavior intact.

The renderer state machine correctly distinguishes prose mentions of triple-backticks from legitimate fence openings. Closed by Phase 7 verify.

Verified algorithmically rather than via model-emitted prose (which is unreliable to elicit deterministically). Three unit cases prove the N1 SOL/post-newline anchor at commit 11d0e59: A. `use ``` to denote code in markdown.` -> filter does NOT trigger (no preceding newline; mid-line backticks ignored). Output passes through verbatim, no [HL:..] highlight marker. B. `intro\n```py\nx=1\n```` -> filter DOES trigger after the \n (post-newline anchor). [HL:py]x=1[/HL] emitted. C. `prose with ``` mention.\nthen:\n```lua\nx=2\n```` -> mid-line skipped; real fence after \n still triggers. Mixed-stream behavior intact. The renderer state machine correctly distinguishes prose mentions of triple-backticks from legitimate fence openings. Closed by Phase 7 verify.

claude-noether closed this issue

2026-05-16 22:37:58 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: marfrit/aish#62