hermes-profile-evolution
Route proposed Team Mimir profile behavior changes into target-owned Hermes dialogue after producer-owned custody classification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Route proposed Team Mimir profile behavior changes into target-owned Hermes dialogue after producer-owned custody classification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Compile and review an approved Workflow Contract for Codex V2 graph, profile, write, check, reviewer, fallback, and external-action consistency without launching work.
Explicitly run an operator-approved Infiquetra Workflow Contract as a root-orchestrated Codex V2 DAG with exact profiles, runtime receipts, bounded writes, one independent review, one remediation, and one targeted recheck.
Deep-dive one chosen Infiquetra idea into a right-sized requirements document before planning.
Run a structured Infiquetra code-quality review at the work-to-PR boundary. Reads the merge-base diff, runs a built-vs-planned audit plus judgment-selected review lenses, validates findings, writes a durable review artifact, appends to the work-thread saga, and routes — without mutating code. Triggers on "review this PR", "code review", "check my diff", "pre-PR review", or a /work hand-in before shipping.
Review Infiquetra plans, requirements, and SDLC documents for implementation readiness.
CEO/founder-mode scope and ambition review. Rethink the problem, find the 10x product, challenge premises, expand or cut scope deliberately — across four committed modes (SCOPE EXPANSION, SELECTIVE EXPANSION, HOLD SCOPE, SCOPE REDUCTION). Fires upstream of execution on a /plan artifact, STRATEGY.md, /brainstorm output, or an ad-hoc scope question; produces a scope decision and routes deep rigor back to /doc-review and /code-review in a closed loop. Triggers on "think bigger", "expand scope", "is this ambitious enough", "rethink this", "scope review", or a /ceo-review invocation. Does not implement, record strategy, file issues, or change code.
| name | hermes-profile-evolution |
| description | Route proposed Team Mimir profile behavior changes into target-owned Hermes dialogue after producer-owned custody classification. |
Use this skill when a requested Team Mimir edit may change a profile's repo-safe behavior. The named profile retains autonomy: this path sends a proposal for dialogue and does not authorize or perform a profile mutation.
Put the intended repository-relative paths, plain-language intent, and sanitized evidence references in one bounded JSON object. Do not include credentials, secrets, transcripts, logs, databases, host addresses, provider/model overrides, system prompts, or tool definitions.
Pipe that object to the bundled adapter. Run the command from this skill directory so the repository's proven skill-relative install layout resolves the bundled script:
printf '%s' '{"schema_version":1,"intent":"Consider clarifying your review preference.","paths":["profiles/brokkr/SOUL.md"],"evidence_references":["docs/team/README.md"]}' \
| python3 ../../scripts/profile_request.py suggest brokkr
If Team Mimir classifies every path as ordinary repository work, continue through normal review; Hermes is not contacted.
If the result starts Hermes dialogue, retain the returned proposal and continuity data in the
chat. Continue interactively with reply or resume; do not turn the exchange into an offline
queue.
Treat an adapter error as a stop. Fix only the request shape or unavailable producer/service boundary named by the error. Never bypass classification by recreating its policy locally.
reply and resume accept bounded JSON on standard input. A reply object has exactly
schema_version, proposal, and message; a resume object has exactly schema_version and
proposal. The proposal is the canonical envelope previously sent to Hermes.
The Codex hook covers recognized edits through apply_patch, Edit, and Write. It is advisory
beyond those supported tool calls and does not claim to prevent shell commands, external editors,
disabled hooks, untrusted hooks, or same-user/root activity. Hook trust is an operator decision.