Context before the change.
Evidence after it.

Step 1 Index the repository offline. No agent wiring with --agent none.

$ repo-ai init --agent none
[  0.01s] repo-ai 0.9.0.dev4 — initialising …/demo
[  0.02s] scanning …/demo
[  0.03s] scan complete: 25 indexable files (30 seen, 6 skipped, 0 unchanged)
[  0.03s] parsing 25 files (stdlib ast + regex fallback)
[  0.03s] found 5 manifests, 10 declared dependencies
[  0.44s] git: main@72990b0, 1 commits, 1 contributors
[  0.45s] analysis complete in 0.4s
 
  repo-ai 0.9.0.dev4 — demo: 25 files, 212 LOC, 2 component(s) [Python, TypeScript]
  agents: none (docs only)   learning: suggest   skills: curated   layout: .repo-ai/context/
  changed 13 file(s)
    create  .gitignore
    create  .repo-ai/config.toml
    (+ 11 generated docs under .repo-ai/context/)
  rules: 0 active, 3 proposed (`repo-ai learn` to review)
  ! tree-sitter not installed: non-Python symbol extraction uses the regex fallback. Install with `pip install 'repo-ai-cli[ast]'` for higher fidelity.

Step 2 A ranked shortlist for a task, with the reason for each file.

$ repo-ai context "fix the login validation bug"
<repo-ai-context>
Retrieved by repo-ai for this request. Confidence: high. Task class: auth, bug. Start with these files; ignore the rest of the repository unless they prove insufficient.
 
Verify with: `make test`
 
| # | File | Why | Look at |
|---|------|-----|---------|
| 1 | `app/auth/login.py` — Login flow: credential validation and token issuance. | keyword match, named like the task, auth path, imported by test_login.py | L1-24 (class LoginError, function validate_credentials) |
| 2 | `tests/test_login.py` | keyword match, named like the task, auth path, imports login.py | whole file (8 lines) |
| 3 | `web/src/components/LoginForm.tsx` | keyword match, named like the task, auth path, imported by LoginForm.test.tsx | L1-10 (function LoginForm) |
… rows 4–9 and the @-mentions line trimmed
</repo-ai-context>

Step 3 Read-only plan: what the change touched and which tests can show a break.

$ repo-ai verify
repo-ai verify: 1 changed file against HEAD
 
changed
  app/auth/login.py  [source, modified, units: validate_credentials]
      evidence: 1 linked test file(s); runtime evidence none
 
selected 1 of 3 test files
  tests/test_login.py  [static] imports app/auth/login.py
 
confidence: MEDIUM: some of the change rests on static evidence only (imports or file names), not on runtime evidence
… uncertainty notes trimmed
run
  targeted:   python -m pytest -q -p no:cacheprovider tests/test_login.py
  full suite: python -m pytest -q tests   (before merging)
  sandboxed:  repo-ai verify --run   (--full runs the full suite instead)
  sharper:    repo-ai verify --record   (once: records which code each test runs)

Step 4 Run the selection sandboxed, in a disposable copy; the tree is checked before and after.

$ repo-ai verify --run
verifying 1 selected test file(s) in the seatbelt sandbox, in a disposable copy of the working tree…
  command: make test  [declared in the project's manifests, CI or docs]
  isolated copy: 31 files (clone, 0.0s)
… strategy line trimmed
  pass    tests/test_login.py (0.4s)
… 3 summary lines trimmed
 
PASSED: every selected test passed (the full suite was not run)
  1 passed
  source tree unchanged
  ! no runtime evidence: selection rests on static imports only (`repo-ai verify --record` records which code each test executes)
… one more caveat trimmed
evidence saved; `repo-ai evidence` shows whether it still covers the working tree.
This covers only the tests that ran; it is not proof the change is correct.

Step 5 Plan a mission offline. No provider, no tests, no writes.

$ repo-ai do "Reject emails with surrounding whitespace in login validation" --dry-run
ASK_APPROVAL: External execution requires verified provider policy, explicit disclosure and budget authorization; planning only; trivial change classification refused unnecessary specialist fan-out
Planning only; no provider or tests invoked.
  app/auth/login.py: keyword match; named like the task; auth path
  web/src/components/LoginForm.tsx: keyword match; named like the task; auth path
  app/api/users.py: keyword match; auth path; imports login.py
… 3 more files trimmed
Unknowns: static import graph only: dynamic imports, reflection, dependency injection, plugin registries and string-based routing are not visible
Orchestration (trivial): merge_owner=controller

Step 6 Which agent CLIs are on PATH. Static check only; nothing is started.

$ repo-ai agents list
codex: PLANNING_ONLY | version not probed | auth UNKNOWN
claude: PLANNING_ONLY | version not probed | auth UNKNOWN
gemini: PLANNING_ONLY | version not probed | auth UNKNOWN
bob: PLANNING_ONLY | version not probed | auth UNKNOWN
agy: PLANNING_ONLY | version not probed | auth UNKNOWN
No model was invoked. Installed executable does not establish authenticated runnable availability.