Tools
Last updated 12 September 2026
| Tool | When | What comes back | Next |
|---|---|---|---|
hexum_skill | First Hexum use this session if SKILL.md is missing, or when intensity changes. | Skill file body, write path, whether to update architecture.hexum. | Write body to path if missing. Follow that file. |
hexum_shrink | Before reading a repo or a wide file set. | A smaller view of those files, plus token counts for the call. | Read that view. Do not dump the originals into the prompt. |
hexum_expand | When one folded function, class, or method is the thing to patch. | The full source of that symbol. | Edit from that body. Do not expand the whole file just in case. |
hexum_splice | When applying the patch. | The updated region, still valid source. | Keep the write small: one symbol or one line. You keep Git. |
hexum_check | Before claiming the turn is done. | healthy, drift_detected, or ARCHITECTURE_CONFIG_MISSING, plus violations. | On drift, follow hint and check again. Do not ship a red check. |
hexum_review | When reviewing a git diff or a GitHub pull request. | approved or changes_requested, violations, token delta, GitHub markdown. | Post the markdown, or let the GitHub webhook do it. |
hexum_control_status | When the agent is repeating edits or failing the same check. | Drift score, health, churn, suggested forbid lines. | If stuck_in_loop, write the patch before more edits. |
hexum_loop_start | When starting a gated build loop for a goal. | loop_id and the next instruction (shrink, edit, check, test, tick). | Follow the instruction. Tick with compile and test output. |
FAQ
What are the five Hexum products?
Hexum is an ecosystem: five products and twelve tools on one key. Efficient: don’t paste the whole repo. Arch: block a bad import. Review: architecture review on GitHub PRs. Control: when the agent loops, write the rule. Loop: only keep building if checks and tests pass.
How does Hexum cut what I pay the model?
Your agent stops dumping the whole repo into the prompt. Hexum sends this task, not every file. On the published card, MBPP context dropped 85.2% and the combined 24-task cut is 87.6%, with 24/24 tests still passing. Tokens are ceil(utf8_bytes/4), not a provider invoice. See /benchmarks.
Does the architecture check use an AI?
No. It reads a rules file in your repo and reports forbidden imports. It does not call a model, so it does not guess.