Measure Goose startup context before the ordinary-skill refactor

coilyco-ops commented

2026-07-23 17:54:35 +00:00

Member

Parent and purpose

Parent: AOS #655.
This ticket produces a fast structural before-and-after baseline for AOS #629.
Scope is deliberately one open-source harness: Goose. The ordinary-skill refactor follows immediately, so building adapters for every selected lane would measure a payload that is about to change.

Fixed measurement lane

Harness: goose.
Role: ops.
Intent: operational-decision.
Default repository and CWD: the AOS repository root. The command accepts an explicit repository and CWD for reproducible fixtures and later reruns.
Delivery path: the committed AOS provider plus the existing Goose composition wiring.

The measurement is structural. It never calls inference and never queries AOSH, backend models, hardware, endpoints, or live services. This ticket does not depend on AOS #650 or the complete selected harness board.

Acceptance

Measure the effective eager startup payload for the fixed Goose lane:
- global composed doctrine
- Goose-specific load points and overlays
- the effective root and nested AGENTS.md cascade for the declared repository and CWD
- role-eager context
- ordinary and selected COMPOSED.md skill frontmatter
- harness and plugin-contributed skill frontmatter
- eager MCP tool schemas delivered to Goose
Keep lazy skill bodies, lazy MCP servers, runtime retrieval, and peripheral substrate files separate from eager startup cost.
Emit a machine-readable inventory and a readable summary with component class, source owner, source path, delivery path, eager or lazy class, bytes, token count or documented token proxy, and payload hash.
Record one before snapshot, then rerun the same lane with the same declared inputs after AOS #629. Report the component and total deltas.
Keep output ordering and payload hashes deterministic.
Fail clearly when the Goose load point, fixed role, fixed intent, or composed provider input is missing or malformed.
Add fixture-backed tests for the Goose adapter, nested AGENTS.md cascade, eager-versus-lazy classification, stable ordering, and before-versus-after delta rendering.
Update the context-budget walkthrough for this narrow baseline. This internal measurement does not add a FEATURES.md entry.

Non-goals

No 16-lane report.
No adapter framework for every selected harness or product surface.
No cross-harness comparison or cross-harness budget gate.
No behavioral or instruction-adherence evaluation. AOS #653 remains separate.
No model, hardware, endpoint, or live-service lookup.

A broader matrix can be reconsidered after the skill refactor only if the Goose before-and-after result shows that another harness would change a concrete decision.

## Parent and purpose * Parent: AOS #655. * This ticket produces a fast structural before-and-after baseline for AOS #629. * Scope is deliberately one open-source harness: Goose. The ordinary-skill refactor follows immediately, so building adapters for every selected lane would measure a payload that is about to change. ## Fixed measurement lane * Harness: `goose`. * Role: `ops`. * Intent: `operational-decision`. * Default repository and CWD: the AOS repository root. The command accepts an explicit repository and CWD for reproducible fixtures and later reruns. * Delivery path: the committed AOS provider plus the existing Goose composition wiring. The measurement is structural. It never calls inference and never queries AOSH, backend models, hardware, endpoints, or live services. This ticket does not depend on AOS #650 or the complete selected harness board. ## Acceptance * Measure the effective eager startup payload for the fixed Goose lane: * global composed doctrine * Goose-specific load points and overlays * the effective root and nested `AGENTS.md` cascade for the declared repository and CWD * role-eager context * ordinary and selected `COMPOSED.md` skill frontmatter * harness and plugin-contributed skill frontmatter * eager MCP tool schemas delivered to Goose * Keep lazy skill bodies, lazy MCP servers, runtime retrieval, and peripheral substrate files separate from eager startup cost. * Emit a machine-readable inventory and a readable summary with component class, source owner, source path, delivery path, eager or lazy class, bytes, token count or documented token proxy, and payload hash. * Record one before snapshot, then rerun the same lane with the same declared inputs after AOS #629. Report the component and total deltas. * Keep output ordering and payload hashes deterministic. * Fail clearly when the Goose load point, fixed role, fixed intent, or composed provider input is missing or malformed. * Add fixture-backed tests for the Goose adapter, nested `AGENTS.md` cascade, eager-versus-lazy classification, stable ordering, and before-versus-after delta rendering. * Update the context-budget walkthrough for this narrow baseline. This internal measurement does not add a `FEATURES.md` entry. ## Non-goals * No 16-lane report. * No adapter framework for every selected harness or product surface. * No cross-harness comparison or cross-harness budget gate. * No behavioral or instruction-adherence evaluation. AOS #653 remains separate. * No model, hardware, endpoint, or live-service lookup. A broader matrix can be reconsidered after the skill refactor only if the Goose before-and-after result shows that another harness would change a concrete decision.

coilyco-ops referenced this issue

2026-07-23 17:55:18 +00:00

Inventory effective AGENTS.md cascades across substrate and product repos #657

coilyco-ops referenced this issue

2026-07-23 17:55:59 +00:00

Enforce generated repo-pointer presence for every eligible managed repo #658

coilyco-ops referenced this issue

2026-07-23 17:57:23 +00:00

[epic] Bound composed-agent context across selected roles and harnesses #655

coilyco-ops referenced this issue

2026-07-23 17:57:37 +00:00

Sync the confirmed AOSH role-intent harness board into AOS #650

coilyco-ops referenced this issue

2026-07-23 17:57:39 +00:00

Measure fine-grained local composed-role instruction adherence #653

coilyco-ops changed title from ~~Measure startup context for every selected role-intent-harness lane~~ to Measure Goose startup context before the ordinary-skill refactor

2026-07-24 00:40:27 +00:00

coilyco-ops referenced this issue

2026-07-24 00:41:21 +00:00

Inventory effective AGENTS.md cascades across substrate and product repos #657

coilyco-ops commented

2026-07-24 00:54:54 +00:00

Author

Member

Parallel structural work from #657 landed on main at 5c822c3. The lane-budget implementation can import discover_repositories, load_scenarios, and active_cascade from agentic_os.agents_context_inventory instead of reimplementing fleet resolution, AOSH separation, nested AGENTS discovery, bridge/override attribution, or ordered source-set hashing. This commit did not edit check_context_budget.py. The infrastructure manifest must use owner-qualified refs because the legacy bare agentic-os entry is ambiguous on a host with multiple checkouts of that name.

Parallel structural work from #657 landed on main at `5c822c3`. The lane-budget implementation can import `discover_repositories`, `load_scenarios`, and `active_cascade` from `agentic_os.agents_context_inventory` instead of reimplementing fleet resolution, AOSH separation, nested AGENTS discovery, bridge/override attribution, or ordered source-set hashing. This commit did not edit `check_context_budget.py`. The infrastructure manifest must use owner-qualified refs because the legacy bare `agentic-os` entry is ambiguous on a host with multiple checkouts of that name.

coilyco-ops referenced this issue from a commit

2026-07-24 01:06:21 +00:00

feat(context): add Goose refactor baseline

coilyco-ops closed this issue

2026-07-24 01:06:21 +00:00

coilyco-ops referenced this issue

2026-07-24 01:09:29 +00:00

Group the Goose baseline and serialize it as YAML #682