Add hidden experts CLI command

[suhaanthayyil]

https://entire.io/gh/entireio/cli/trails/709

What is this?

Adds a hidden/labs CLI command: entire experts.

Before you edit an unfamiliar part of a repo, it answers:

“Which agents already have real evidence for this code — and what did they use?”

It is agent/workflow provenance, not a human popularity list. The output ranks prior Codex / Claude / Cursor (etc.) sessions by evidence: matching checkpoints, agent-attributed lines, skills invoked, tools used, MCP servers, and files touched — without exposing raw prompts or chat transcripts.

Discover it via entire labs (listed as experimental).

---

Example usage

Directory scope (most common — “who knows this package?”):

entire experts cmd/entire/cli/

Single file:

entire experts cmd/entire/cli/experts_cmd.go

Explicit repo (owner/repo resolves to a repo ULID via your accessible repo list; ULIDs also accepted):

entire experts cmd/entire/cli/ --repo entireio/cli --limit 5

Staged files (what am I about to commit?):

entire experts --staged

Interactive browse (TTY only — two-pane viewer, session links open on entire.io):

entire experts cmd/entire/cli/ --tui

JSON (for scripts / agents):

entire experts cmd/entire/cli/ --json

Natural-language query (requires code-search index on the API cell; otherwise 503 with a setup hint):

entire experts "stripe webhook retry logic" --repo entireio/cli

---

Example output (plain text)

Agent provenance for entireio/cli (main)

Codex
  evidence: 1 sessions, 1 matching checkpoints, 9 steps, 142 agent-attributed lines
  skills: go-cli (2)
  tools: shell (4), search (3)
  mcp: github (1)
  files: cmd/entire/cli/experts_cmd.go
  - feat: experts provenance  - 1 checkpoints, 9 steps

Claude Code
  evidence: 1 sessions, 1 matching checkpoints, 14 steps, 210 agent-attributed lines
  skills: go-cli (1), tui (1)
  tools: fileOps (6)
  mcp: github (2)
  files: cmd/entire/cli/experts_cmd.go, cmd/entire/cli/experts_tui.go
  - TUI polish  - 1 checkpoints, 14 steps

Session names are clickable links (OSC 8) to https://entire.io/gh/{owner}/{repo}/session/{id} when running in a capable terminal.

---

How it works (short)

1. You pass a path scope, staged files, or a natural-language query.
2. The CLI resolves --repo (git origin, owner/repo, or repo ULID) and calls the data API:
   • GET /api/v1/repos — resolve owner/repo → ULID (authz-safe)
   • POST /api/v1/repos/{id}/experts — rank agent provenance for the scope
3. The API (now on entire-api, merged in entirehq/entire-api#126) searches checkpoint/session history for file matches and returns residency-safe summaries.
4. The CLI renders ranked profiles (plain, --json, or --tui).

Privacy: JSON/plain/TUI only surface agent evidence fields (sessions, counts, skills, tools, matched files). No prompts, emails, or human author identity.

---

What changed in this PR

• Hidden top-level experts command + entire labs listing
• Flags: --repo, --branch, --limit, --json, --staged, --tui, hidden --insecure-http-auth
• Scope handling: local paths (incl. from subdirs), deleted path-like scopes, absolute paths in worktree, --staged additions/renames/deletions
• Retarget from legacy cache route to entire-api POST /api/v1/repos/{id}/experts
• TUI: two-pane agent list + evidence detail; j/k navigate, tab sections, enter expand sessions, o open session, q quit
• Data API auth: loopback HTTP discovery for local dev when --insecure-http-auth is set

---

Screenshots

TUI (--tui)

---

Verification

• GitHub checks on head d0c782b5d: license, lint, binary-size, test-core, test-canary, test-integration (a/b/c), test — all passed
• Trail 709: 0 open findings
• go test ./cmd/entire/cli -run Experts -count=1

Companion

• API (merged): entirehq/entire-api#126
• Supersedes retired entire.io cache route PR

[Copilot]

Pull request overview

Adds a new hidden/labs entire experts command to query and render agent/skill/tool provenance for either path scopes (including staged files) or natural-language queries, with optional JSON output.

Changes:

• Register a hidden top-level experts command and advertise it via entire labs.
• Implement request construction (repo resolution, scope vs query detection, staged scopes) and render JSON or agent-centered text output.
• Add unit tests covering query vs scope behavior, staged scope collection (including deletions), and output constraints (no human-identity fields in JSON).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
cmd/entire/cli/root.go	Registers the hidden experts command on the root command.
cmd/entire/cli/labs.go	Adds experts to the experimental command listing and labs overview text.
cmd/entire/cli/experts_cmd.go	Implements the experts command: flags, request shaping, API call, and rendering.
cmd/entire/cli/experts_test.go	Adds tests for command wiring, scope/query behavior, staged scopes, and output format constraints.

[suhaanthayyil]

Retargeted to the entire-api data plane (entirehq/entire-api#126).

entire experts now calls POST /api/v1/repos/{id}/experts on the per-cell entire-api service instead of /api/v1/cache/:org/:repo/experts. --repo accepts a repo ULID (used directly) or an owner/repo, resolved to its ULID via the caller's accessible-repo list (GET /api/v1/repos) -- authz-safe, no existence leak. TUI, palette, session links, and staged/query routing are unchanged.

Verified: go build, go vet, and go test ./cmd/entire/cli pass on the committed tree.