Survey: Tmuxinator as Potential Session Initializer for Miadi/Hermes Navigator

Parent issue: jgwill/Miadi#334 Related: jgwill/Miadi#265 — Hermes Navigator v1 focus contract PDE: bcd9a22d-9b1c-4c69-a659-8610b3663769 Artefact path: /a/src/IAIP/prototypes/artefacts/survey--miadi-session-initializer--TmuxinatorAsPotentialSessionInitializer--pde-2605171739--bcd9a22d-9b1c-4c69-a659-8610b3663769/ Session: c51ee7ec-de59-4ba1-bb2f-2858d2601af7

---

Recommendation (headline)

Do not adopt Tmuxinator. Extend hnavig scene activate instead.

Hermes Navigator already owns named sessions (hnavig session new/attach), workspace scenes (hnavig scene activate/snapshot/resume), and raw subprocess tmux integration. Tmuxinator would duplicate this orchestration and add a Ruby/gem dependency to a Python toolchain.

---

Research Tracks (4 parallel agents)

Agent	Coverage
web-tmux-survey	Himmelwright article (2019) + Tmuxinator profile + fit assessment
hnavig-capability-scan	Full Hermes Navigator session/scene/tmux capability map
branch-mission-context	Branch mission, PDE, rispecs, issue anchors, IAIP AGENTS.md
academic-context	6 academic fields, 20+ authors, 5 Montreal researchers, 3 thesis framings

---

Key Findings

1. Himmelwright's article is a plain shell tutorial, not a Tmuxinator endorsement. Its idempotency guard (tmux list-sessions | grep ) is trivially embeddable in Python.
2. Hermes Navigator already has hnavig scene activate with multi-window workspace support, SQLite session persistence, and audit logging.
3. The real gap is branch-aware declarative workspace profiles (per-branch TOML startup recipes), not Tmuxinator adoption.
4. Tmuxinator's value is as an optional export format only (human-editable YAML recipes). Not a core dependency.
5. No academic paper studies Tmuxinator empirically — this is a research gap and opportunity.

---

Academic Context

6 fields: Programmer Tools HCI, Empirical SE (DevOps/IaC), LLM-Based AI Agents, Human-in-the-Loop AI, Cognitive Load/DevX, Declarative IaC.

Montreal researchers (thesis directors):

• Bram Adams (Polytechnique Montréal, MCIS Lab) — most directly relevant; release engineering + IaC configuration
• Foutse Khomh (Polytechnique Montréal, SWAT Lab) — MLOps/LLMOps pipelines
• Martin P. Robillard (McGill) — developer comprehension, API usability
• Joëlle Pineau (McGill + MILA) — reproducible AI, RL
• Weiyi Shang (Concordia) — log engineering, system observability

3 defensible thesis directions documented in SURVEY-REPORT.md, grounded in Yang et al. SWE-agent ACI (2024), Xi et al. LLM Agents Survey (2023), and Meyer et al. DevX research (FSE 2014, 340 citations).

---

Artefact Files

• AGENTS.md — original mission brief (IAIP side)
• SURVEY-REPORT.md — full synthesis report (created 2026-05-17)

---

HITL Ambiguities (require human judgment)

1. Is #334's deliverable this report or an implemented hnavig scene extension?
2. Scope: Hermes Navigator only, or all branch hosts (gaia, as, eury)?
3. Profile format: YAML vs. TOML vs. Python dict?
4. Is Tmuxinator-export worth the maintenance surface?