conduit · skills used in this session, from the session bar

Ask: show how many skills a session has used in its header, and on click or tap, the list. That includes skills you invoked (/brainstorming in the composer) and skills the agent loaded itself (a Skill tool call).

Every mock uses the same session: 48 minutes old, 5 skills, 7 loads. You ran 2, the agent ran 3, one ran 3 times across both of you, and one is loading right now.

Shared rules (every design)

Count = distinct skills"5", not "7 loads". Repeats show as ×3 on the row. A skill is one idea; loading it twice isn't two skills.

Slash means youYour skills look like the composer pill (/name, pink). Agent skills get the sparkle in indigo. One rule, already learned from the composer.

Tap a row, land on itJumps the transcript to where it ran and opens that turn's existing skills toggle. Repeated skills jump to the latest; long-press or right-click picks an earlier one or opens its SKILL.md.

Hidden at zeroNo "0 skills". The control appears with the first load, like the unread and attention badges do.

LiveA skill loading now pulses until the call finishes, so you can see the agent reach for one mid-turn.

No new storage (probably)Both sources are already in the event store: /skill tokens in user messages (tokenizeSkills) and Skill tool calls (skillName). Needs a check: built-in commands like /compact must not count.

1 · Count chip

recommended

A small ✦ 5 chip beside the title, built from the same pill as the Unread and Settled chips. Tap opens a bottom sheet on phones and a popover on desktop: one row per skill, oldest first.

Header cost: ~44px of width on the title row. The title truncates first, as it already does for the state chip.
Collapsed bar: Hidden, same as the state chip. Scroll up to get it back.
Taps to list: 1.
Build cost: Low. Reuses Badge, Menu (sheet / popover) and the turn skills toggle's jump.

Pick this unless you want the names visible without a tap. It's the smallest change that fully answers the ask, and it follows every pattern the bar already uses.

2 · Pill strip

names at a glance

The skills themselves sit under the title as pills, newest first, cut off by +N. Tap a pill to jump to it. Tap +N (or the strip's end) for the same full list as design 1.

Header cost: +28px of height on phones, while any skill exists. On desktop it eats the bar's empty middle, and long names start truncating the title.
Collapsed bar: Hidden.
Taps to list: 0 for the newest 2–3, 1 for the rest.
Build cost: Low–medium. Overflow measuring for +N is the fiddly part.

Use it if you glance at which skills are active often enough to want it without tapping. The price is a permanently taller phone bar, which design 20 of the bar worked hard to avoid.

3 · Skills view

room to read

Skills become a fifth view next to Chat, Terminal, Diff and Files. The list is grouped by turn, and desktop gets a side pane showing the selected skill's SKILL.md.

Header cost: None on phones. The count lives in the Views menu.
Collapsed bar: Unchanged.
Taps to list: 2 on phones (Views → Skills).
Build cost: Medium. A new view plus a SKILL.md reader, though GetSkillContent already exists.

Falls short of the ask on phones: the count isn't in the header until you open Views. Its real value is the SKILL.md pane, which could ship later as the target of design 1's "Open SKILL.md".

4 · Timeline ribbon

wild

A hairline along the bottom of the bar is the whole session's timeline. Each skill load is a dot at the moment it happened, pink for you and indigo for the agent. Tap the ribbon and it drops open into a labelled timeline with the list below.

Header cost: +22px of height, everywhere.
Collapsed bar: Could survive as the dots alone, with no label. The only design that can.
Taps to list: 1.
Build cost: Medium. Time scale, dot collisions when skills load seconds apart.

Shows something the others can't: when in the session the skills clustered (planning up front, debugging in the middle, review at the end). Dense sessions turn the dots into mush, and it's a second strip right after the per-turn activity strip.

Compare

My pick: 1. If you want more, add 3's SKILL.md pane as the target of 1's "Open SKILL.md" later. Pick 2 instead only if you'd use the names at a glance.

Design	Count in header	Names without tapping	Phone height	Shows when	Build
1 Count chip	Yes	No	+0	Per row	Low
2 Pill strip	Yes (+N)	Newest 2–3	+28px	Order only	Low–med
3 Skills view	Desktop only	No	+0	By turn	Medium
4 Ribbon	Yes	No	+22px	On a timeline	Medium

Open questions. (a) Count distinct skills (5) or loads (7)? These mocks say distinct. (b) Should skills loaded inside subagents count toward the parent session? These mocks say no. They'd show on the subagent's own session.