summarizing-friction-logs

Summarizing friction logs

Collect every frog friction entry in a workspace into one read-only dashboard, so known friction can be prioritized in a single view.

Rules

• Read-only. Never run frog resolve or frog publish, never edit

entries, and never install frog.

• Never drop entries silently. Anything not rendered is counted.
• `pending` matters most. A pending entry is filed nowhere else, so it

is invisible outside this dashboard; a linked one already has an issue.

Find the logs

The workspace is the directory that holds the user's repositories: the path the user gives, otherwise the parent of the current repository's main clone (from a linked worktree, the first path git worktree list prints). Outside a repository, ask once.

find <workspace> -maxdepth 3 -type d -path '*/.agents/friction-log' -not -path '*/.claude/worktrees/*'

Worktrees are excluded because they share the primary clone's log and would double-count entries. <workspace>/.agents/friction-log also matches: it is not a repo but the global store for system-wide and cross-repo friction (a CLI used everywhere, machine setup, an agent harness). Label it global, never the workspace's directory name, and never count it as a repo.

Collect entries

frog list --cwd <scope> --format json

<scope> is the discovered path minus /.agents/friction-log. Entries carry id, title, severity (blocker, major, minor), state (pending or linked), and an optional target (upstream owner/repo). Without frog, read each <scope>/.agents/friction-log/<id>/friction.md; its frontmatter carries title, severity, and target.

Count each scope before reading detail. A scope with more than 15 entries is bounded: sort by severity, then newest first, render only the top 10, and add an aggregate line for the rest, for example + 31 more (4 major, 27 minor) not rendered.

For each rendered entry, read its friction.md and distill the body into a one-line gist: what got in the way, plus the suggested fix when it names one.

Dashboard

One Markdown dashboard, in this shape:

• Header: totals with global counted separately, for example

7 entries across 3 repos + 2 global: 1 blocker · 3 major · 5 minor, plus any bounded scopes, for example (bounded: global 41→10).

• Table, sorted by severity, then newest first. Glyphs: 🔴 blocker,

🟠 major, 🟡 minor. Logged comes from the id's YYYYMMDDHHMMSS prefix. Append (upstream: owner/repo) to the title when an entry has a target.

• Detail: each entry's gist, grouped by scope, global last.
• Footer: scopes with zero entries, logs that could not be read, and one

line per bounded scope with its total and the unrendered counts.