[Birfy]

Summary

Completes the Kind filter + fidelity addendum on #384 (does not close the whole issue — MethodPolicy catalog / remaining port migrations stay open).

• Every registry entry is labeled surface=card in list / show / MCP / plan echo.
• ACE / GEPA / DGM / SkillOpt / EvoSkill cards carry research_port + fidelity_note so same name ≠ paper port.
• Host /algorithms slash + SKILL.md filter by artifact kind and state the card≠port rule.
• Catalog + port-fidelity docs updated; --kind / MCP kind= already refused unsupported cards in plan.

Test plan

• pytest tests/test_algorithms.py (incl. new fidelity tests)
• pytest tests/test_slash_command_parity.py
• CI required checks on the tip

Refs #384

[Birfy]

Verdict: PASS
Tip SHA: d58e315
Ran: pytest tests/test_algorithms.py (full file, incl. new fidelity tests) + tests/test_slash_command_parity.py on the tip; risk_gate.py --pr 810 → T3 (paths agentdescent/integrations/SKILL.md, agentdescent/integrations/__init__.py)
Not run: waiting on required CI (build + test 3.9–3.13) at review time

Correctness

• Algorithm.describe() always sets surface: "card"; ACE/GEPA/DGM/SkillOpt/EvoSkill also get research_port + an honest fidelity_note that names the fidelity class and examples/<name>/. RRSI and other cards without a sibling port correctly omit those keys.
• CLI list/show, MCP list_algorithms/describe_algorithm, and plan's _algorithm_echo all carry the same surface / port fields so hosts cannot read a mechanism slice as the paper port.
• Kind filtering is reinforced (host slash + SKILL + MCP tool text + empty-kind note); policy_slot still gets its dedicated note and does not clobber into a false “cards vs ports” story when the list is empty for that kind.
• New tests lock the contract: test_card_vs_research_port_fidelity_fields, test_cli_show_and_plan_echo_card_surface, test_mcp_list_algorithms_carries_cards_vs_ports_note (incl. kind filter excluding evoskill from kind=text).

Design fit (#384 addendum only)

• Matches the Kind filter + Card vs port acceptance addendum on plugin: let users select/run algorithms from examples/ via EvolveSpec / CLI / MCP #384 without pretending to close MethodPolicy catalog / remaining port migrations — PR body and changelog say so clearly.
• Docs (algorithms.md via gen script, port-fidelity.md, algorithm-authoring.md) and host copy (SKILL.md, /algorithms command) stay in sync with the registry frozenset _CARDS_WITH_RESEARCH_PORT.

Nits (non-blocking)

• _CARDS_WITH_RESEARCH_PORT is a manual frozenset; fine for five names, but a drift test against examples/<name>/ (or against docs/algo-*.md) would catch a future same-name card that forgets the port link.
• Generated research-port bullets in docs/algorithms.md repeat the full fidelity_note string — readable but a bit long; optional later tighten.

Merge

risk_gate T3 (integrations host surface). Per policy: do not auto-merge; waiting on Danyang's click / 「都合了」 after required CI is green on this tip. AGENT_PAUSE=false.

[Birfy]

Self-review (Birfy)

Tip SHA: d58e315
Ran:

• pytest -q tests/test_algorithms.py tests/test_slash_command_parity.py → pass
• python -m tools.gen_algorithms_docs --check → pass
• python -m tools.changelog_fragments check → pass
• risk_gate → T3 (integrations/SKILL.md + integrations/init.py)
  Not run:
• full pytest -q suite — CI covers 3.9–3.13
• Ark real-model run — no model path changed
  Findings:
• none. Completes plugin: let users select/run algorithms from examples/ via EvolveSpec / CLI / MCP #384 kind-filter + fidelity addendum only; MethodPolicy catalog remains open on plugin: let users select/run algorithms from examples/ via EvolveSpec / CLI / MCP #384.
  Verdict: PASS

[Birfy]

Self-review (Birfy)

Tip SHA: cd44b7f
Ran:

• pytest -q tests/test_integrations.py::test_checked_in_claude_plugin_matches_the_package tests/test_integrations.py::test_checked_in_dsh_plugin_matches_the_package tests/test_slash_command_parity.py tests/test_algorithms.py → pass
  Not run:
• full suite — CI re-run on this tip
• Ark real-model run — no model path changed
  Findings:
• none (prior tip failed only on checked-in plugin SKILL drift)
  Verdict: PASS

[Birfy]

Verdict: PASS
Tip SHA: cd44b7f
Ran: pytest tests/test_algorithms.py tests/test_slash_command_parity.py tests/test_integrations.py::test_checked_in_claude_plugin_matches_the_package tests/test_integrations.py::test_checked_in_dsh_plugin_matches_the_package; risk_gate → T3 (integrations/SKILL.md + init.py + host plugin paths)
Not run: full pytest -q (defer to required CI)

Note

Tip advanced from d58e315 → cd44b7f (plugin parity sync); prior PASS was on the old tip so this re-PASS matches current head.

Risk: T3 — waiting for Danyang's merge click after CI green. No --auto.

[Birfy]

Self-review (Birfy)

Tip SHA: cd44b7f
Ran:

• pytest -q tests/test_integrations.py::test_checked_in_claude_plugin_matches_the_package tests/test_integrations.py::test_checked_in_dsh_plugin_matches_the_package tests/test_slash_command_parity.py tests/test_algorithms.py → pass
• python -m tools.gen_algorithms_docs --check → pass
• python -m tools.changelog_fragments check → pass
• CI build + test 3.10–3.13 → green (3.9 still finishing)
  Not run:
• Ark real-model run — no model path changed
  Findings:
• none
  Verdict: PASS

[Birfy]

Verdict: PASS
Tip SHA: b4e0768
Ran: merged origin/main (#814) resolving docs/port-fidelity.md (kept #384 surface note + decision-tree link); pytest tests/test_algorithms.py tests/test_slash_command_parity.py tests/test_changelog_fragments.py; risk_gate → T3
Not run: full pytest -q (CI re-runs 3.9–3.13)

Risk: T3 — waiting for Danyang's merge click after CI green. No --auto.

[Birfy]

Conflict after #812 landed on main (e510e07). Tip b4e07684 is now CONFLICTING with agentdescent/integrations/SKILL.md / checked-in plugin assets.

Please rebase onto current main, re-sync generated plugin/SKILL files if needed, and re-post PASS on the new tip. Required CI must be green again before merge (T3; user already clicked 「合」 once this tip is ready).

[Birfy]

Verdict: PASS
Tip SHA: 2b8958f

Rebased onto main fc24d929 (after #812 Path A + #813 factory Policies) with force-with-lease. Dropped the earlier merge commit; history is two commits again.

Conflict resolution:

• docs/port-fidelity.md: kept both the plugin: let users select/run algorithms from examples/ via EvolveSpec / CLI / MCP #384 product-surface note and the Docs+plan: decision tree for card vs MethodPolicy vs full port vs aggregator_factory (no silent Policies drop) #805 decision-tree link (same order as the prior merge tip).
• integrations/dsh-agentdescent/lib/index.js: regenerated via render_dsh_plugin() from current agentdescent/integrations/SKILL.md so Path A wording and card≠port fidelity both land (no hand-merge of the CONTENT string).

Ran:

• pytest -q tests/test_algorithms.py tests/test_slash_command_parity.py tests/test_changelog_fragments.py → pass
• pytest -q tests/test_integrations.py::test_checked_in_claude_plugin_matches_the_package tests/test_integrations.py::test_checked_in_dsh_plugin_matches_the_package → pass
• python -m tools.gen_algorithms_docs --check → pass
• python -m tools.changelog_fragments check → pass
• risk_gate.py --pr 810 → T3 (integrations/SKILL.md, integrations/init.py, host plugin paths)

Not run: full pytest -q (required CI 3.9–3.13 re-running on this tip)

Risk: T3 — no --auto. Mergeable again; squash-merge only after required CI is green and under Danyang's existing 「合」 for this PR.