Followups & Lineage

--followup chains a new run onto an existing OpenAI or Azure Responses API session. The model keeps its prior reasoning context; you supply additional prompt + files. Lineage is tracked in oracle status so you can see the whole tree at a glance.

#Why followup instead of starting fresh

• Cheaper. You don't re-pay for the original input tokens.
• Coherent. The model remembers its earlier conclusions and can reason about changes.
• Auditable. oracle status shows the parent/child tree.

#Basic flow

# Initial run
oracle --model gpt-5.5-pro --slug arch-review \
  -p "Audit the auth flow end-to-end" \
  --file "src/auth/**"

# Later — continue with new files
oracle --followup arch-review \
  -p "Re-evaluate now that the rate-limiter is wired in." \
  --file "src/auth/rate-limiter.ts"

--followup accepts:

• A stored session id (a1b2c3…)
• A session slug (arch-review)
• An OpenAI / Azure response id (resp_abc1234…) — useful for chaining onto runs that didn't originate in Oracle.

#Multi-model parents

When the parent used --models a,b,c, pick which lineage to continue from with --followup-model:

oracle --followup arch-review --followup-model gpt-5.5-pro \
  -p "Continue from the Pro answer" \
  --file "src/auth/rate-limiter.ts"

Without --followup-model, Oracle errors with the available lineage.

#What's chainable

If you try to follow up on an unsupported provider, Oracle errors clearly instead of silently starting fresh.

#Browser multi-turn

In browser mode, --browser-follow-up adds extra prompts to the same ChatGPT conversation, which is the closest equivalent to a chained API followup:

oracle --engine browser --model gpt-5.5-pro \
  -p "Review this migration plan" --file docs/migration.md \
  --browser-follow-up "Challenge your previous recommendation" \
  --browser-follow-up "Give the final decision"

Each --browser-follow-up is sent after the previous turn completes. Not supported in Deep Research mode.

#Lineage in oracle status

Status    Model         Mode    Timestamp           Chars    Cost  Slug
completed gpt-5.5-pro   api     05/06 09:00 AM      1800  $2.110  arch-review
completed gpt-5.5-pro   api     05/06 09:14 AM      2200  $2.980  ├─ arch-review-rate-limiter
running   gpt-5.5-pro   api     05/06 09:22 AM      1400       -  │  └─ arch-review-implementation
pending   gpt-5.5-pro   api     05/06 09:25 AM       900       -  └─ arch-review-risk-check

Children inherit the parent's slug prefix unless you pass --slug explicitly.

#Common patterns

• Plan → challenge → final. Three turns: ask Pro to plan, follow up with "find the weakest assumption," follow up again with "given the above, give the final plan."
• Bug → repro → fix. Turn 1 sends the failing test + error. Turn 2 sends the suspected file. Turn 3 sends the proposed fix and asks for review.
• Architecture → implementation. Parent run does design; children focus on individual modules. The tree in oracle status becomes the audit trail.

#Limitations

• Followups don't move between providers. You can't follow up an OpenAI run with a Gemini one — open a new session and re-bundle.
• previous_response_id retention on OpenAI / Azure varies by tier. If a followup fails with "response not found," the parent has aged out — start fresh.
• Custom --base-url proxies (LiteLLM, etc.) often strip the response id. Test once before relying on it.