cross-backend-porting-EXPERIMENTAL

[Enforcer03]

Summary by cubic

Adds cross-backend porting between nexau and mini-swe-agent with an honest coverage audit and report; introduces hdp.engine.port (audit() and port()) plus a port CLI/subcommand.

• New Features

  • hdp.engine.port: audit(doc, dest_target) enumerates coverage gaps; port(source → dest) runs lift → audit → generate, writes hdp_port_coverage.json, and refuses on blocking issues unless allow_partial.
  • Coverage model: CapabilityIssue and CoverageReport with severities blocking, silent_drop, collision.
  • Adapter support:
    • nexau: flags blocking policy/non-external verifier, unresolvable tool bindings, and destination collisions.
    • mini-swe-agent: flags blocking policy/non-external verifier, unresolvable tool bindings, and system_rules ids with no manifest slot as silent_drop.
  • CLI: python -m hdp.engine.port ... and hdp.engine.run port; config via hdp.port block (see configs/hdp/port-example.yaml). Scope is nexau ↔ mini-swe-agent (openharness excluded).
  • Tests: added $0 e2e and adapter coverage tests ensuring honest reporting and refusal behavior.

• Refactors

  • Adapters expose capability_issues(doc); base.FrameworkAdapter provides a no-op default.
  • _reject_unsupported routes through capability_issues but preserves behavior: policy/verifier remain NotImplementedError; unresolvable tool bindings still raise ValueError during generate().
  • STATUS.md updated with Phase 6 port readiness.

^{Written for commit 491055f. Summary will update on new commits.}

[cubic-dev-ai]

2 issues found across 11 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="hdp/engine/run.py">

<violation number="1" location="hdp/engine/run.py:154">
P2: Config-driven port runs can unexpectedly allow partial generation when `allow_partial` is supplied via `${ENV}` as `false`, because `bool('false')` evaluates to `True`. Consider parsing string booleans explicitly before passing this into `port_harness`, so blocking coverage gaps still refuse by default.</violation>
</file>

<file name="hdp/engine/port/coverage.py">

<violation number="1" location="hdp/engine/port/coverage.py:28">
P2: The `severity` field on `CapabilityIssue` is typed as `str` but documented and universally used as one of three exact values (`"blocking"`, `"silent_drop"`, `"collision"`). The project already uses `Literal` for this pattern (`ChangeKind = Literal["added", "removed", "modified"]` in `differ.py`). Using `Literal["blocking", "silent_drop", "collision"]` would catch typos at type-check time and keep the codebase consistent.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: Config-driven port runs can unexpectedly allow partial generation when allow_partial is supplied via ${ENV} as false, because bool('false') evaluates to True. Consider parsing string booleans explicitly before passing this into port_harness, so blocking coverage gaps still refuse by default.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At hdp/engine/run.py, line 154:

<comment>Config-driven port runs can unexpectedly allow partial generation when `allow_partial` is supplied via `${ENV}` as `false`, because `bool('false')` evaluates to `True`. Consider parsing string booleans explicitly before passing this into `port_harness`, so blocking coverage gaps still refuse by default.</comment>

<file context>
@@ -140,6 +140,42 @@ def cmd_gen(cfg: dict) -> int:
+    harness_path = port_cfg.get("harness", hdp_cfg.get("harness", "agents/code_agent_simple"))
+    source_target = port_cfg.get("source_target", "nexau")
+    dest_target = port_cfg.get("dest_target", "mini-swe-agent")
+    allow_partial = bool(port_cfg.get("allow_partial", False))
+    arm = (cfg.get("run") or {}).get("arm", "treatment")
+    seed = int((cfg.get("run") or {}).get("seed", 0))
</file context>

Suggested change

	allow_partial = bool(port_cfg.get("allow_partial", False))
	raw_allow_partial = port_cfg.get("allow_partial", False)
	allow_partial = (raw_allow_partial.strip().lower() in {"1", "true", "yes", "on"}
	if isinstance(raw_allow_partial, str) else bool(raw_allow_partial))

[cubic-dev-ai]

P2: The severity field on CapabilityIssue is typed as str but documented and universally used as one of three exact values ("blocking", "silent_drop", "collision"). The project already uses Literal for this pattern (ChangeKind = Literal["added", "removed", "modified"] in differ.py). Using Literal["blocking", "silent_drop", "collision"] would catch typos at type-check time and keep the codebase consistent.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At hdp/engine/port/coverage.py, line 28:

<comment>The `severity` field on `CapabilityIssue` is typed as `str` but documented and universally used as one of three exact values (`"blocking"`, `"silent_drop"`, `"collision"`). The project already uses `Literal` for this pattern (`ChangeKind = Literal["added", "removed", "modified"]` in `differ.py`). Using `Literal["blocking", "silent_drop", "collision"]` would catch typos at type-check time and keep the codebase consistent.</comment>

<file context>
@@ -0,0 +1,45 @@
+    component_id: str
+    layer: str
+    type: str
+    severity: str   # "blocking" | "silent_drop" | "collision"
+    reason: str
+
</file context>