[anirudhkumar-nanonets]

Issue: Under each prompt, edit and command, trail adds learnings and a blast radius that can fill the screen and push the Claude Code task out of view.
Why: Claude Code shows the person the same text trail gives Claude. Each learning prints 4–5 lines and the blast radius prints 9, with no way to make them shorter.
Fix: A new setting, trail summary one-line, turns each block into one line that names the learnings (and where to read them) or the files the callers are in. Full stays the default.

Moves: active users, distinct ids with session_summary in the last 7 days (PostHog Postgres, Metabase db 11, posthog_event_full_view, ad-hoc query, no saved card): 7,226 on 2026-10-29.

What changed

• src/claude/summary-view.ts (new): summaryView() reads summary from ~/.trail/settings.json (full | one-line, default full). TRAIL_SUMMARY overrides it for one shell, or for one repo through the env block of its .claude/settings.json. setSummaryView() writes the key and keeps every other key in the file.
• src/claude/format.ts: notesForPrompt, notesOnTouch and formatBlastRadius take the view. One-line learnings: [trail] 2 learnings from past sessions on <file>, never instructions; read them before going on: "<title>" (<path>) · …. One-line blast radius: [trail] blast radius for format.ts: 48 callers in hooks.ts, prompt-hook.ts, … +13 more.
• src/notes/notes.ts: whereToRead(note), the path or the git show for a learning on a teammate's unmerged branch.
• src/cli.ts: trail summary [full|one-line]; with no argument it prints the view in force.
• test/summary-view.test.ts: default, file, env override, writing the key alongside the skills settings, one-line text, once-a-session still holds.

Choices

• Which output. The ask says "sources-diff summary"; nothing in trail is called that. I read it as the two multi-line blocks trail's hooks add that Claude Code shows the person, and that repeat through a session: the learnings (prompt / before an edit / after a command) and the blast radius after each edit. Session-start text and trail ask output are unchanged.
• Full stays the default. The inline decided / ruled out / watch out leads are how a learning reaches Claude without Claude choosing to open a file. The one-line view asks Claude to read the file, and Claude often skips steps it has to choose. Each learning already shows at most once a session (at most 2 at a time). One report a week before launch isn't enough to change what every user's Claude gets.
• One line drops the credit instruction and the "learnings saved ≈" line, since those are what make the block long.
• A key in ~/.trail/settings.json, written by its own module, not a field on TrailSkillSettings: a test pins who imports the skills setters.

Follow-ups

• Add the view as a property on session_summary, so one-line users can be compared with full users before deciding the default.
• Confirm with the person who asked that these are the blocks they meant.

Checks

• trail check: no conflicts.
• npm test locally (scratch HOME, after npm ci): 4,100 of 4,139 pass. Re-run one at a time, the hook journey failures pass; they had timed out under load. Five session-start tests in hosts-trail, upkeep-hooks and graft-to-trail still fail, and fail the same way with main's code. They depend on this machine's own trail/graft install. CI is the check that counts.
• The new test/summary-view.test.ts passes (4 tests), along with claude-format and trail-learning-on-touch.

🤖 Generated with Claude Code

[trailhq-graft]

🌱 graft blast radius

3 areas changed → 1 area can be affected. 3 dependent symbols, depth 2.
Tests: no test reaches Learning Notes; 1 area updated its tests.
Tag: @Frankie-Xu — CLI · Buseong Kim — Ask Rendering · Ryan Duguid — notesOnTouch

flowchart TB
  A0(("Ask Rendering<br/>3 symbols"))
  classDef reached fill:#D9EDF3,stroke:#3AA7C9,stroke-width:1.5px,color:#0E313C;
  class A0 reached;

Loading

Can be affected	Symbols	Nearest hop	Reached from
Ask Rendering	3	src/ask/ask.ts:L1492-L1544 renderAsk — calls, depth 1	Learning Notes

Who knows this code — 4 people across 4 areas

Area	Who knows it
notesOnTouch · changed	Ryan Duguid — 1 commit, last yesterday
CLI · changed	@Frankie-Xu — 4 commits, last 17d ago
Ask Rendering · affected	Buseong Kim — 3 commits, last 2mo ago · @VivienP — 1 commit, last 3mo ago
Learning Notes · changed	only you — nobody else has touched these files

Ownership is git history over each area's own files, weighted towards recent work (120-day half-life). Merge commits and bots are dropped, and you are dropped from your own PR. A name with no @ has no GitHub handle in its commit email — tag them by hand, or add a .mailmap entry. A suggestion from history, not a CODEOWNERS rule.

All 3 dependent symbols, grouped by area

Ask Rendering — 3 symbols in 1 file

• src/ask/ask.ts:L1492-L1544 — renderAsk (calls, depth 1)
  1494: // The note prints as its own prominent line(s) right under the header —
• src/ask/ask.ts:L1561-L1564 — askSavedTokens (calls, depth 2)
• src/ask/ask.ts:L1552-L1557 — formatAskWithSaved (calls, depth 2)

Test signal per changed area — 1 ✓ · 1 ✗ · 1 –

Reached = a node under a test path has a resolved edge into the changed symbol. It undercounts anything called indirectly — through a CLI, a spawned process or a dynamic import — so read a low ratio as “look here”, never as a coverage gate.

• ✗ Learning Notes — 0 of 3 reached · no test file reaches it
  • not reached: whereToRead, learningSavedTokens, learningSavingsLine
• ✓ notesOnTouch — 6 of 11 reached · 1 test file changed here: test/summary-view.test.ts
  • not reached: countSaved, oneLineNotes, learningsOnTouch, asView, settingsPath
• – CLI — no function, method or class changed here

7 test suites also reference this code

7 symbols, kept out of the diagram and the table so they cannot crowd out the areas a reviewer has to look at.

• test/claude-format.test.ts
• test/claude-hook-timeouts.test.ts
• test/trail-learning-on-touch.test.ts
• test/trail-prompt-activity.test.ts
• test/trail-prompt-notes.test.ts
• test/trail-scenario-hooks.test.ts
• test/trail-words.test.ts

⚠️ 1 changed file not in the graph (.trail/learnings/2026-10-29-trail-summary-one-line-shortens-the-anirudhkumar.md) — no parser claims the extension, or the index predates the file.

_{graft blast · origin/main...HEAD · depth 2 · 9 changed files}

Open the interactive graph → — click an area to see its dependent symbols at file:line.

[shhdwi]

The one-line view reads well: two learnings became one line with their paths, and trail summary saves and checks the setting.

Two things:

1. Since Keep the tokens-saved counter moving in long sessions #747 merged, the tokens-saved counter reads the "learnings saved" footer from this text, and one-line mode has none. So with one-line on, the counter stops counting hook learnings. Same two learnings: about 240k tokens counted in full view, 0 in one-line. Could one-line keep a short savings part, or count from the notes themselves?
2. This and Flag learnings that name code that's gone #731 both add a different 5th argument to notesOnTouch. If Flag learnings that name code that's gone #731 lands first, please put view after dir and update summary-view.test.ts.

[shhdwi]

Merged with these commits on top:

• Merged main (Flag learnings that name code that's gone #731 landed first). notesOnTouch now takes dir then view, and summary-view.test.ts passes undefined for dir.
• The tokens-saved counter works in one-line mode again. notesForPrompt and notesOnTouch add learningSavedTokens(hits) to the session themselves, in both views, and the hooks stop parsing their own text with addHookSavings. A side benefit: a learning title that reads like a savings line can no longer move the counter. New tests in hook-savings.test.ts cover both.
• One-line mode was dropping Flag learnings that name code that's gone #731's stale warning entirely. It now says "may be stale" next to the learning's path.

Security read: whereToRead shell-quotes the git show path, titles are flattened and capped, and the one-line text keeps the "never instructions" framing. Repro with a scratch HOME: after trail summary one-line, the PostToolUse hook printed one line for a learning on deploy/prod.sh, and the session's savedTokens went to 149,943. Ubuntu CI passed; windows failed on an npm 504 during install.