Agentic invariant-mining toolkit for P (propose → check → repair → judge)#974
Closed
ankushdesai wants to merge 1 commit into
Closed
Agentic invariant-mining toolkit for P (propose → check → repair → judge)#974ankushdesai wants to merge 1 commit into
ankushdesai wants to merge 1 commit into
Conversation
…→ judge) Standalone tooling that drives `p compile` / `p check` to mine and verify candidate specification monitors for P programs. An LLM proposes intent-level invariants from the source; PChecker (the sound oracle) verifies every one; the agent then judges each result as verified / vacuous / bug / spurious. No dependency on the PInfer C# backend — these are pure utility scripts: - check_candidates.py : wire candidate monitors into a project, generate one test each, model-check, classify HOLDS / FAILS(+cex) / VACUOUS, with vacuity canaries and detection of pre-existing SUT failures (reported INCONCLUSIVE, not a false bug). - prep_candidates.py : deterministic P fixups for LLM-generated specs (entity unescape, var-decl hoisting, `!in`/`for` rewrites). - diag_compile.py / repair_workflow.js / verify_repairs.py : compile-and-repair loop that feeds the exact compiler error back to an LLM to fix non-compiling candidates (recovered all dropped candidates in two rounds on the tutorials). - README_agentic_invariants.md : how the pipeline fits together. Portable dotnet/`p` resolution (Linux + macOS). Experimental; intended for discussion alongside the PInfer effort. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Pull request overview
Adds a standalone, agent-oriented “propose → check → repair → judge” utility toolkit under Src/PInfer/Scripts/ to mechanically wire LLM-proposed P spec monitors into existing P tutorial projects, compile/check them with PChecker, diagnose compilation failures, and support iterative LLM repair.
Changes:
- Introduces
check_candidates.pyto inject candidate specs + per-candidate tests into a P project, runp compile/p check, and classify results (HOLDS/FAILS/VACUOUS/INCONCLUSIVE). - Adds compile-diagnosis + repair support via
diag_compile.py,repair_workflow.js, andverify_repairs.py. - Adds deterministic candidate cleanup with
prep_candidates.pyand end-to-end documentation inREADME_agentic_invariants.md.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 8 comments.
Show a summary per file
| File | Description |
|---|---|
| Src/PInfer/Scripts/check_candidates.py | Wires candidate specs/tests into a project and runs PChecker to classify outcomes, including vacuity canaries. |
| Src/PInfer/Scripts/diag_compile.py | Compiles candidates one-at-a-time to capture the first compiler error per spec for repair. |
| Src/PInfer/Scripts/prep_candidates.py | Applies deterministic transformations to LLM-generated P (unescape, !in rewrite, for→foreach, var hoisting). |
| Src/PInfer/Scripts/repair_workflow.js | Workflow wrapper to LLM-repair non-compiling monitors using compiler errors + syntax rules. |
| Src/PInfer/Scripts/verify_repairs.py | Recompiles repaired specs to measure recovery rate and report remaining failures. |
| Src/PInfer/Scripts/README_agentic_invariants.md | Documents the overall pipeline and how to run the scripts. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| import html, json, re, sys | ||
| from pathlib import Path | ||
|
|
||
| HANDLER_RE = re.compile(r'(on\s+\w+\s+do\s*\([^)]*\)\s*|on\s+\w+\s+goto\s+\w+\s+with\s*\([^)]*\)\s*|entry\s*(?:\([^)]*\))?\s*|exit\s*)\{') |
Comment on lines
+101
to
+105
| def wire(subset): | ||
| cand_dst.write_text("\n".join(block_of[n] for n in subset)) | ||
| test_dst.write_text("// AUTO-GENERATED\n\n" + "\n".join( | ||
| f"test tc_{n} [main={args.main}]:\n assert {n} in {args.assert_in};\n" for n in subset)) | ||
| return "Compilation succeeded" in run(["p", "compile"], proj, env).stdout |
Comment on lines
+111
to
+113
| out = run(["p", "check", "-tc", f"tc_{n}", "-i", str(args.iters)], proj, env).stdout | ||
| m = BUG_RE.search(out) | ||
| bugs = int(m.group(1)) if m else -1 |
| for i in range(len(spans) - 1): | ||
| name, block = spans[i][1], src[spans[i][0]:spans[i + 1][0]].strip() | ||
| diag.write_text(block) | ||
| out = subprocess.run(["p", "compile"], cwd=d, env=env, capture_output=True, text=True).stdout |
Comment on lines
+31
to
+33
| def main(): | ||
| env = build_env() | ||
| data = json.loads(Path(sys.argv[1]).read_text()) |
| d = r["dir"] | ||
| diag = Path(d) / "PSpec" / "_diag.p" | ||
| diag.write_text(r["fixedSource"]) | ||
| out = subprocess.run(["p", "compile"], cwd=d, env=env, capture_output=True, text=True).stdout |
Comment on lines
+49
to
+50
| ).then(r => ({ benchmark: f.benchmark, name: f.name, changes: r && r.changes, fixedSource: r && r.fixedSource })) | ||
| )).then(a => a.filter(Boolean)) |
Comment on lines
+67
to
+70
| def main(): | ||
| data = json.loads(Path(sys.argv[1]).read_text()) | ||
| by = data.get('result', data) | ||
| outdir = Path(sys.argv[2]) if len(sys.argv) > 2 else Path('/tmp') |
Member
Author
|
Superseded by #977, which carries the hardened successors of these scripts (frozen candidate schema, auto-canary vacuity, static name gate, baseline-differential attribution, ranking) plus the design plan (Src/PInfer/Scripts/PLAN.md). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Standalone tooling that uses an LLM to propose intent-level specification invariants for P programs, has PChecker verify every one (the sound oracle), and then judges each result. No dependency on the PInfer C# backend — these are pure utility scripts driving
p compile/p check.The loop:
Files (
Src/PInfer/Scripts/)check_candidates.pyINCONCLUSIVE(not a false bug)prep_candidates.pyvar-decl hoisting,!in/forrewrites)diag_compile.pyrepair_workflow.jsverify_repairs.pyREADME_agentic_invariants.mdWhy this shape
<Name>_canarycompanion reveals a monitor whose assertion never fires.Validation (4 tutorial benchmarks)
Scope / status
Experimental utility tooling, portable dotnet/
presolution (Linux + macOS). Opened for discussion alongside the PInfer effort; the propose/judge steps are LLM-driven (run with Claude Code or any agent), this PR is the deterministic middle.🤖 Generated with Claude Code