Omh Commit PR Authoring

Catalog Metadata

Category: verification Phase: commit-pr-authoring Hermes role: reviewer Quality tier: evidence-ledger-gated Reasoning demand: standard

Quality bar:

• Read the template and at least the last few commits before writing a word, and record what was found.
• Build the evidence ledger first; the Tested: and validation sections are projections of it, not prose.
• Write the subject in the repository's shape and the body as why the change exists, not a restated diff.
• Load references/commit-and-pr-conventions.md for trailer, sign-off, and template-section mapping instead of improvising them.
• Put Closes #N on its own line only when every success criterion of that issue is met.

Handoff policy:

Keep the convention read, the drafted text, and the evidence ledger in Hermes. The commit, the push, and opening the PR are executor or user actions; OMH never performs them, and a command reaches Tested: only from an observed run record.

Required inputs:

• the change: the diff, the staged files, or the branch compared with its base
• the repository's PR template and a sample of its recent commit log
• the commands that ran for this change, each with its observed exit status
• the commands that were prepared, planned, or skipped, and why
• the issue or goal the change closes, when there is one

Expected outputs:

• repo_convention_read/v1
• evidence_ledger/v1
• commit_message_draft/v1
• pr_body_draft/v1 when a PR is being opened

Artifact expectations:

• repo_convention_read/v1 names the template path, the subject shape and trailers seen in the recent log, and the sign-off rule, or says which of them was not found
• evidence_ledger/v1 lists every command with observed or not_observed; observed carries the exit status and where it was seen
• commit_message_draft/v1 and pr_body_draft/v1 take Tested: and the validation section only from observed ledger rows

Safety rules:

• List a command under Tested: only when the evidence ledger records it as observed; a command that was prepared, planned, or described as passing without a run record goes under Not-tested:, whatever the wording says.
• Never commit, amend, push, or open the PR; OMH prepares the text and the user or the executor records it.
• Read the repository's own template and recent log before drafting; impose no convention the repository does not use.
• Do not claim CI, review, or merge state in the text unless it was observed, and name what was not.
• Never put credentials, tokens, private URLs, or raw transcripts into a commit message or PR body.

Runtime Evidence

Record observed delegation results; otherwise return not_available or not_observed. Prepared OMH routing is not execution, review, CI, merge-readiness, or merge evidence.

• Treat wrapper memory/context summaries as advisory local context, not proof of opaque Hermes memory reads or changes. Preserve workflow intent and stop conditions; verify before claiming completion. Reply in the user's own words and the host's own voice: its SOUL.md persona owns reply language, tone, speech level, and sentence endings, progress updates included (where it sets no language, use the one the user wrote in), and OMH shapes structure and content only; OMH's record terms (surface, lane, wrapper, handoff, evidence boundary, not_observed) stay in records and tool calls, never in the sentence the user reads unless they ask about one; and when a stop condition or a decision the user owns ends the turn, offer the next action as a question rather than declaring what will not be done.

Use Hermes-native subagent/delegation features when available: native subagents -> Hermes delegation when available, otherwise sequential lanes.

Shared product, compatibility, topology, memory, harness, and execution rules: omh-routing/references/skill-common-rail.md. Load it when applicable; otherwise name an unavailable capability.