fix(cli): respect explicit --max-turns value even when it equals default - #87
Merged
Merged
Conversation
max_turns used 60 as both the default and the sentinel to detect whether the user passed the flag. This meant `--max-turns 60` was indistinguishable from "not passed", so the env var HERMES_MAX_ITERATIONS would silently override the explicit CLI value. Change the default to None so any user-supplied value takes priority.
RUFFY-369
added a commit
to RUFFY-369/hermes-agent
that referenced
this pull request
Apr 12, 2026
angelburgosrosado
pushed a commit
to angelburgosrosado/hermes-agent
that referenced
this pull request
Apr 27, 2026
…inel fix(cli): respect explicit --max-turns value even when it equals default
This was referenced May 4, 2026
Brecht-H
pushed a commit
to Brecht-H/hermes-agent
that referenced
this pull request
May 5, 2026
Adds a `skill_invocations` table to state.db and writes one row per skill listed on a cron job at completion (success or failure paths). Tokens, cost and duration are sourced from the existing session row. Includes a `skill_stats_daily` view that buckets invocations by day and (skill_name, model, provider), and a new `SessionDB.query_skill_ema()` method that applies exponential weighting (default alpha=0.3, ~5d half-life) so the dashboard can A/B local Qwen against external models once analyzer crons start firing. SCHEMA_VERSION 11 → 12. Pure additive: existing rows untouched, new table created on next connection-open via the existing executescript() path. No Alembic. Slash-command and ad-hoc skill_view invocations are NOT tracked in v1. Multi-skill crons over-account: each skill in `job["skills"]` gets the full session cost. Both are acceptable for the analyzer-cron use case (1 skill per cron) and can iterate later. Constraint: no new external dependencies — uses sqlite3 + stdlib only. Rejected: per-skill cost split (would require model attribution inside a single agent run, which Hermes does not currently track) | Reason: defer to v2 once NousResearch#87 surfaces real-world skew. Confidence: high (smoke-tested end-to-end on tmp DB) Scope-risk: narrow (additive table, no existing-row touchpoints) Not-tested: live cron fire (validated by Step 3 — Pass 2 v2 handover) Machine: orion-terminal Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
someaka
pushed a commit
to someaka/hermes-agent
that referenced
this pull request
May 24, 2026
test_interrupt_during_child_api_call_detected was flaking on Python 3.11 CI runners taking 1.4-1.5s to detect interrupt. Bumped threshold to 2.0s so the test passes under CI load. Fork CI run: Tests NousResearch#87 on commit d2e76b1b28
someaka
pushed a commit
to someaka/hermes-agent
that referenced
this pull request
May 30, 2026
test_interrupt_during_child_api_call_detected was flaking on Python 3.11 CI runners taking 1.4-1.5s to detect interrupt. Bumped threshold to 2.0s so the test passes under CI load. Fork CI run: Tests NousResearch#87 on commit d2e76b1b28
someaka
pushed a commit
to someaka/hermes-agent
that referenced
this pull request
May 31, 2026
test_interrupt_during_child_api_call_detected was flaking on Python 3.11 CI runners taking 1.4-1.5s to detect interrupt. Bumped threshold to 2.0s so the test passes under CI load. Fork CI run: Tests NousResearch#87 on commit d2e76b1b28
someaka
pushed a commit
to someaka/hermes-agent
that referenced
this pull request
Jun 9, 2026
test_interrupt_during_child_api_call_detected was flaking on Python 3.11 CI runners taking 1.4-1.5s to detect interrupt. Bumped threshold to 2.0s so the test passes under CI load. Fork CI run: Tests NousResearch#87 on commit d2e76b1b28
Meraniya
pushed a commit
to Meraniya/hermes-agent
that referenced
this pull request
Aug 6, 2026
Companion piece to hermes-already-has-routines.md — documents how the bundled llm-wiki skill implements Karpathy's LLM Wiki pattern (gist.github.com/karpathy/442a6bf555914893e9891c11519de94f), element by element, plus where Hermes extends it (provenance markers, source-drift hashing, scheduled ingestion, multi-platform querying, headless Obsidian sync). Claude-Session: https://claude.ai/code/session_01Kk61nsX85RGno5HYGyFkQZ Co-authored-by: Claude <noreply@anthropic.com>
sijav
added a commit
to sijav/sijav-agent
that referenced
this pull request
Aug 7, 2026
…al-pool routing tests/agent/test_credential_pool_routing.py::TestFailureAttribution:: test_unmatched_key_does_not_retry_only_pool_entry fails on this branch and is NOT from my work — verified failing at the parent of my first commit today. Filed with the mechanism and an explicit warning to determine whether the code regressed or the test went stale before changing either.
sijav
added a commit
to sijav/sijav-agent
that referenced
this pull request
Aug 7, 2026
…le-entry guard dead code The credential-pool sole-entry guard's comment states exactly the failing test's rationale, but instrumentation shows it never fires: _available_entries returns 2 entries for a 'single-entry' pool because the pool auto-seeds the agent's own runtime key. That also explains why identity matching reports 'matched no entry' while an entry holding the failing key exists. Recorded with the probe output and a candidate fix, flagged as an auth-semantics decision rather than a mechanical test fix.
sijav
added a commit
to sijav/sijav-agent
that referenced
this pull request
Aug 7, 2026
…redentials (NousResearch#87) conftest's autouse credential guard was half-built. read_claude_code_credentials reconciles TWO sources — the macOS Keychain and ~/.claude/.credentials.json — but the guard stubbed only the Keychain. That path is built from Path.home() directly, so it sits outside the HERMES_HOME sandbox, and this conftest deliberately does not redirect HOME ('that broke subprocesses'). On Linux/WSL, where the Keychain never existed, the guard was therefore a complete no-op and every anthropic test read the developer's live OAuth token. Observed damage: a test building an isolated single-entry credential pool actually loaded TWO entries — its own plus a real 'claude_code' one — so credential_pool's sole-entry guard (whose comment states 'a single-entry pool cannot rotate') could never fire. That silently broke pool routing and is why test_unmatched_key_does_not_retry_only_pool_entry has been red since the commit that introduced it: it is environment-dependent, green on CI where no such file exists and red on any developer machine with Claude Code installed. The write side was the bigger hazard: _write_claude_code_credentials targets the same real path and is reachable from the pool's refresh path, so an unguarded run could rewrite live credentials. Guarded with a DENY-LIST rather than a stub — the pattern this conftest already uses for kanban writes. The real reader/writer still run; they refuse only when the resolved path is the developer's actual ~/.claude/.credentials.json, captured at import before any fixture rewires the environment. A blanket stub broke the 12 tests that legitimately redirect HOME and exercise the real reader; the deny-list keeps them working. tests/agent: 534 files, 7428 tests, 0 failed.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
--max-turnsused60as both the default value and the sentinel to detect whether the flag was passed (cli.py:795)CLI arg > env var > config file > default--max-turns 60was indistinguishable from not passing the flag, soHERMES_MAX_ITERATIONSenv var would silently override the explicit CLI valueNoneso any user-supplied value always takes priorityReproduction
HERMES_MAX_ITERATIONS=30python cli.py --max-turns 60max_turnsis set to 30 (env var overrides explicit CLI arg)max_turnsis set to 60 (CLI arg respected)