| name | fable-method-harness |
| description | Operating harness for working on the method-harness-compiler project (or reviewing its artifacts). Use when starting ANY task on that project — phase reviews, tool discovery, PR review, eval design, memory updates, runtime export, repo maintenance, or A/B test design. Routes the agent to a minimal per-task file set instead of whole-repo reading. For work on ANY OTHER project, activate only when the task is long / multi-step, multi-agent, expensive-if-wrong, makes a completion claim (done / passing / fixed / ready / safe / staged), or touches governance (permissions, hooks, cron, CI, routing) — the portable entry is then core/GLOBAL_BOOTSTRAP.md. Skip for one-line / typo / format edits and pure capability asks. |
| id | SKILL |
| layer | entry |
| purpose | Lightweight runtime wrapper — lets skill-aware runtimes (Claude Code and similar) discover and launch the harness like a skill |
| read_when | Auto-discovered by skill-convention runtimes; read manually when wiring the harness into a new runtime |
| depends_on | ["BOOTSTRAP.md","ROUTES.yaml"] |
| used_by | ["ROUTE-phase-review","ROUTE-tool-discovery","ROUTE-pr-review","ROUTE-eval-design","ROUTE-memory-update","ROUTE-runtime-export","ROUTE-repo-maintenance","ROUTE-ab-test-design"] |
| tags | ["entrypoint","skill","runtime-wrapper"] |
| retrieval_keywords | ["skill entrypoint","launch harness","runtime wrapper","when to use"] |
SKILL: fable-method-harness
This file only launches the harness. The rules live in the routed files.
1. When to use this skill
Any task touching the method-harness-compiler project: planning, review,
building, evaluating, releasing, or maintaining. Also when asked to "work
the way this project works."
For large or multi-agent (ultracode) tasks on any OTHER project, the
portable entry is core/GLOBAL_BOOTSTRAP.md (core/ only — skip section 2).
For auditing / simplifying / benchmarking a harness itself (CLAUDE.md,
hooks, skills, slash commands, scheduled reviews, rolling maintenance), the
entry is the adaptive-harness skill adapter:
.claude/skills/adaptive-harness/SKILL.md. This root file launches work ON
the mhc project; the adapter launches work ON a harness — they share the
repo but route to different file sets.
2. Required startup sequence
Read, in order: context/L0_bootstrap.md → context/L2_current_phase.md →
context/L3_task_router.md → your route's bundle in ONE call (python scripts/route_pack.py <task_type>; entry-only fallback:
scripts/route_show.py or grep - id: ROUTE-<...> in ROUTES.yaml — do
not read it whole). (Same ladder as BOOTSTRAP.md.)
3. Route selection
Classify the task into one of the 8 task types (L3 table), load its
bundle via scripts/route_pack.py <task_type> (one call = entry +
start+required contents; entry-only: route_show.py or a grep of the
block), open optional only with a stated reason. Never bulk-load the
repo, never serial-read the route files one by one.
4. Required output style
End every substantive deliverable with: Decision / Risks / Required
changes / Next actions. Phase-gate questions get an explicit
GO / CONDITIONAL_GO / NO_GO.
5. Self-check
Before returning, score the draft 1–5 against the rubric(s) your route
names; fix or disclose anything below 3. Use the matching checklist under
validation/self_checklists/ when one exists.
6. Project memory update
If the work changed direction, phase, or standing decisions: append to the
memory/ files per operating_model/project_memory_policy.yaml. Append,
never overwrite; cite the grounding artifact or commit.
7. What not to do
Do not read the whole repo first. Do not work outside the L2 allowed
list without an explicit gate decision. Do not fabricate citations, data,
or transcripts. Do not execute or install discovered tools. Do not put
secrets here — the repository is PUBLIC (docs/publication_status.md);
every file must stay public-safe.