Skip to content

fix(agent): inject turn-level live time context#10448

Closed
Sapientropic wants to merge 1 commit into
NousResearch:mainfrom
Sapientropic:fix/turn-level-live-time-context
Closed

fix(agent): inject turn-level live time context#10448
Sapientropic wants to merge 1 commit into
NousResearch:mainfrom
Sapientropic:fix/turn-level-live-time-context

Conversation

@Sapientropic

@Sapientropic Sapientropic commented Apr 15, 2026

Copy link
Copy Markdown

Summary

Hermes currently has a session-level Conversation started: ... timestamp, but it does not provide a reliable turn-level live sense of "now" for each API call.

This PR adds a turn-scoped Current time: YYYY-MM-DD HH:MM (TZ) hint at API-call time instead of baking live time into the cached session prompt.

What changed

  • add _build_live_time_context() in run_agent.py
  • add _build_turn_scoped_system_prompt() so live time is injected per API call, not into _build_system_prompt()
  • use that helper in both:
    • run_conversation()
    • _handle_max_iterations()
  • add targeted tests for:
    • cached system prompt stays free of Current time:
    • chat-completions run path injects live time
    • max-iteration summary path injects live time
    • codex responses path injects live time

Why this shape

This is intentionally aligned with the request in #10421:

  • live time is turn-scoped
  • it is injected at API-call time
  • it does not mutate the session-cached system prompt
  • it works for both chat-completions and codex-responses paths

Relation to other work

This PR is intentionally narrower than #10061.

They are adjacent, but not the same fix.

Repro verified

I re-verified the gap locally on:

  • latest stable v0.9.0 released on April 13, 2026
  • current main at 422f2866 on April 16, 2026

In both cases, Hermes still had only the session-start timestamp and no turn-level live Current time: context.

How to test

python -m pytest tests/run_agent/test_run_agent.py -k 'turn_level_live_time or excludes_turn_level or summary_injects'
python -m pytest tests/run_agent/test_run_agent_codex_responses.py -k 'turn_level_live_time'

Refs #10421

@Sapientropic Sapientropic mentioned this pull request Apr 15, 2026
Open
@alt-glitch alt-glitch added type/feature New feature or request P3 Low — cosmetic, nice to have comp/agent Core agent loop, run_agent.py, prompt builder labels Apr 26, 2026
@alt-glitch alt-glitch mentioned this pull request Apr 26, 2026
Open
@quinnmacro

quinnmacro commented Apr 26, 2026

Copy link
Copy Markdown

Hi @Sapientropic — I noticed this is closely related to my PR #15872 which addresses the same root cause (no turn-level live time). I wanted to flag a key architectural difference for the maintainers:

This PR (#10448) injects Current time: into the system prompt via _build_turn_scoped_system_prompt(). This means the system prompt content changes every turn, which breaks prompt cache prefix — the system message sent to the API is no longer identical across turns.

PR #15872 injects Current time: into the user message (same mechanism as pre_llm_call plugin context). This preserves prompt cache — the system prompt stays frozen and identical across turns, while dynamic time goes into the per-turn user message.

The Hermes codebase itself documents this principle (around run_conversation line 9449):

"Context is ALWAYS injected into the user message, never the system prompt. This preserves the prompt cache prefix — the system prompt stays identical across turns so cached tokens are reused."

However, this PR has advantages that #15872 initially lacked (now added):

  • ✅ Test coverage (3 test paths)
  • ✅ _handle_max_iterations() path covered
  • ✅ Codex responses path covered

I have updated #15872 to also cover _handle_max_iterations() with tests and user-message injection. The codex responses path still needs coverage.

Happy to collaborate on a unified approach — the ideal solution would use user-message injection (for cache preservation) with the comprehensive path coverage and tests from this PR.

@alt-glitch alt-glitch mentioned this pull request Apr 27, 2026
Closed
This was referenced Apr 29, 2026
Open
Open
@markojak

markojak commented Apr 29, 2026

Copy link
Copy Markdown

Closing this as superseded by #17459/#17476.

The problem statement is valid, but the selected direction is to consolidate on a single core ephemeral runtime/user-message context path, preferably based on #15872, while keeping volatile current time out of the cached system prompt/cacheable prefix. A turn-scoped system-prompt path risks reintroducing prompt-cache instability or another parallel timestamp mechanism.

Please rebase any still-useful pieces (tests, wording, helper code) into #15872/#17476 rather than pursuing this PR as a standalone path.

@alt-glitch alt-glitch mentioned this pull request May 1, 2026
Open
@alt-glitch alt-glitch mentioned this pull request May 9, 2026
Closed
3 tasks
This was referenced May 18, 2026
Open
Open
Open
@alt-glitch alt-glitch mentioned this pull request May 27, 2026
Open
@Sapientropic Sapientropic closed this by deleting the head repository Jun 4, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

comp/agent Core agent loop, run_agent.py, prompt builder P3 Low — cosmetic, nice to have type/feature New feature or request

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@Sapientropic @quinnmacro @markojak @alt-glitch @web-flow