Generate loadouts

A loadout is a Markdown knowledge artifact for one task, phase, and role.

Modes

• Routing: summaries, paths, and compact metadata
• Full: inline content up to the ordinary advisory token budget

Per-node reading_priority can override the loadout-level mode with full, summary, or pitfalls-only.

CLI

akms loadout task-17 \
  --repo . \
  --phase 2 \
  --tags plasticity return-mapping \
  --role implementer \
  --mode routing \
  --max-depth 2 \
  --available-context 32000

Default output:

knowledge/loadouts/2-task-17-loadout.md

Use --output for an exact repository-relative path.

Python API

from akms.graph.generate_loadout import generate_loadout

content = generate_loadout(
    G=graph,
    ranked_nodes=ranked,
    task_id="task-17",
    phase=2,
    graph_version=graph_version,
    seed_tags=["plasticity", "return-mapping"],
    agent_role="implementer",
    mode="routing",
    available_context=32000,
    config=config,
    output_path="knowledge/loadouts/2-task-17-loadout.md",
    repo_root=".",
)

Required-aware loadouts

When task_knowledge and a resolution_manifest are supplied by the exact resolver, the generator renders:

1. Required knowledge
2. Coactivated knowledge
3. Advisory domain knowledge
4. Pitfall warnings
5. Session history
6. Suggested reading order

Required and coactivated selections are not truncated by the ordinary advisory budget. The manifest fingerprint is recorded in the header.

Retrieval and timestamps

The generator may use the qmd wrapper for scoped content retrieval and records whether the binary was available. It also writes generated_at. Selection and ordering are deterministic, but raw loadout bytes can differ with timestamp or environment-capability metadata.