Proofline v4.5h: PL-CRIT01 and PL-NEXT2 (0.2.0-next.2)

.proofline/config.json

@@ -0,0 +1,3 @@
+{
+  "approvals": ".proofline/approvals"
+}

CHANGELOG.md

@@ -2,17 +2,19 @@
 
 All notable project changes are recorded here. Dates use ISO 8601 (`YYYY-MM-DD`).
 
-## Unreleased
+## 0.2.0-next.2 (2026-10-21)
 
 ### Added
 - `proofline --version` (and `-v`) prints the installed version and exits 0. On a subcommand `--version` is still an unknown flag.
 - The README and the package carry the Proofline mark (`docs/logo.svg`, `docs/logo-white.svg`, PNGs at 512 and 128).
+- The tests column of a claim line takes `crit:<criterion id>` for a task whose proof is a command the gate runs rather than a test file it adds, such as an intake or measurement task. The name holds only when it is a criterion of that task, it ran in this gate, its type is `command`, `metric`, `test_red_green` or `docs`, and it passed. An unknown id, another task's criterion, a `grep`, `file_exists`, `scope` or `wiring` check, or a criterion that did not pass is a claim mismatch that names the reason. Other names are still looked up in the repository, and `-`, `n/a` and `none` still name no test.
 
 ### Fixed
 - `proofline setup` no longer ends in "Next step: proofline setup" when `AGENTKEY_HOME` is unset or `agentkey` is not on PATH. Each refusal says what the variable or the command is for and prints the exact command that provides it (`export AGENTKEY_HOME=...`, `npm install --save-dev @optitransfer/agentkey`). `proofline contract sign` does the same. The README documents both before the short path.
 - The `AGENTS.md` the README tells you to copy no longer trips the `no_verify` rule of a fresh contract on the first `verify`: its "Never" list names the hook-skipping flag in words.
 - The owner-signature test ran agentkey from a path on one laptop. It now runs the `@optitransfer/agentkey` devDependency, and CI installs the lockfile with `npm ci --ignore-scripts` before the self-gate.
 - The repository moved to `github.com/Optitransfer/proofline`; `homepage`, `repository` and `bugs` in package.json point there. The old URL redirects.
+- A `grep` pattern that starts with `(?i)`, `(?s)` or `(?m)`, the way a Python verifier writes flags, runs with those flags. It crashed as an invalid group, so a contract shared with a Python gate was BLOCKED here and PASS there.
 
 ## 0.2.0-next.1 (2026-10-17)
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@optitransfer/proofline",
-  "version": "0.2.0-next.1",
+  "version": "0.2.0-next.2",
   "description": "Independent V&V for coding agents. The agent says done; Proofline re-runs every check against a contract, rejects claims the evidence does not support, and gates the push.",
   "keywords": [
     "ai-agents",

proofline.json

@@ -3832,9 +3832,205 @@
         "If a criterion cannot pass without loosening a threshold, skipping or deleting a test, raising a timeout, widening the scope or editing the contract: STOP and report file:line and the measured value.",
         "If the environment a criterion needs is missing (repository admin, npm owner login, Actions, a second machine): report BLOCKED naming the missing item; never mark it passed by hand."
       ]
+    },
+    {
+      "id": "PL-CRIT01",
+      "title": "A claim can name an executed criterion with crit:<id>, and a grep pattern may start with (?i) or (?s)",
+      "phase": "implement",
+      "owner": "lane-a",
+      "depends_on": [],
+      "requirements": [
+        "REQ-008",
+        "REQ-009"
+      ],
+      "scope": {
+        "allow": [
+          "src/claims.mjs",
+          "src/engine.mjs",
+          "src/checks.mjs",
+          "test/claims_crit.test.mjs",
+          "test/grep_inline_flags.test.mjs",
+          "CHANGELOG.md"
+        ]
+      },
+      "description": "Two gaps found on 2026-10-20. (1) checkClaim (src/claims.mjs:40-46) accepts a tests-column name only when git grep -F or git ls-files finds it, and drops '-', 'n/a' and 'none', so an intake or measurement task, whose proof is a command the gate runs and not a test file it adds, cannot make an honest done claim (the oll4com/spaceapp #64 intake task PR64-00 had to be recorded PASS-BY-EVIDENCE). The tests column takes crit:<criterion id>; it holds only when the id is a criterion of the claimed task that ran in this gate, of type command, metric, test_red_green or docs, with status pass. An unknown id, another task's criterion, a grep, file_exists, scope or wiring check, or a criterion that did not pass is a claim mismatch naming the reason. The gate passes checkClaim the id, type and status of each criterion it ran (src/engine.mjs). Any other name is still looked up in the repository, and '-', 'n/a' and 'none' still name no test. (2) the grep verifier (src/checks.mjs:211-214) builds new RegExp(v.pattern, v.flags || 'm'); a leading (?i) or (?s), as a Python verifier writes flags, throws 'Invalid group', so the x402-conformance contract (XC-00-AC03, XC-04-AC12, XC-04-AC13, XC-M2-AC02) is BLOCKED here and PASS in its own CI. A leading (?i), (?s) or (?m) becomes flags added to the criterion's own; a pattern without one is unchanged and stays case-sensitive. Both are recorded under Unreleased in CHANGELOG.md.",
+      "criteria": [
+        {
+          "id": "PL-CRIT01-AC01",
+          "mandatory": true,
+          "statement": "crit:<id> holds for a passing executed criterion of the task and is refused, with the reason, for an unknown id, another task, a check that runs nothing and a failing criterion",
+          "verifier": {
+            "type": "test_red_green",
+            "test_file": "test/claims_crit.test.mjs",
+            "cmd": "node --test {test}"
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC02",
+          "mandatory": true,
+          "statement": "a grep pattern starting with (?i) or (?s) runs with those flags instead of crashing, and a pattern without one stays case-sensitive",
+          "verifier": {
+            "type": "test_red_green",
+            "test_file": "test/grep_inline_flags.test.mjs",
+            "cmd": "node --test {test}"
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC03",
+          "mandatory": true,
+          "statement": "Full suite exits 0, nothing hung",
+          "verifier": {
+            "type": "command",
+            "cmd": "npm test",
+            "expect_exit": 0,
+            "timeout_s": 1800
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC04",
+          "mandatory": true,
+          "statement": "the changelog records the crit:<id> claims form under Unreleased",
+          "verifier": {
+            "type": "grep",
+            "file": "CHANGELOG.md",
+            "pattern": "^- The tests column of a claim line takes `crit:<criterion id>`",
+            "must": "present",
+            "flags": "m"
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC05",
+          "mandatory": true,
+          "statement": "the changelog records the inline-flag grep fix",
+          "verifier": {
+            "type": "grep",
+            "file": "CHANGELOG.md",
+            "pattern": "^- A `grep` pattern that starts with `\\(\\?i\\)`",
+            "must": "present",
+            "flags": "m"
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC06",
+          "mandatory": true,
+          "statement": "crit: names are resolved on the default gate path",
+          "verifier": {
+            "type": "wiring",
+            "entry": "bin/proofline.mjs",
+            "symbol": "critIssues"
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC07",
+          "mandatory": true,
+          "statement": "Changes stay in scope",
+          "verifier": {
+            "type": "scope"
+          }
+        },
+        {
+          "id": "PL-CRIT01-AC08",
+          "mandatory": true,
+          "statement": "Committed and pushed to main",
+          "verifier": {
+            "type": "pushed"
+          }
+        }
+      ],
+      "stop_rules": [
+        "If a criterion cannot pass without loosening a threshold, skipping or deleting a test, raising a timeout, widening the scope or editing the contract: STOP and report file:line and the measured value.",
+        "If the environment a criterion needs is missing (repository admin, npm owner login, Actions, a second machine): report BLOCKED naming the missing item; never mark it passed by hand."
+      ]
+    },
+    {
+      "id": "PL-NEXT2",
+      "title": "Version 0.2.0-next.2, so npm carries PL-CRIT01",
+      "phase": "release",
+      "owner": "lane-a",
+      "depends_on": [
+        "PL-CRIT01"
+      ],
+      "requirements": [
+        "PL-R01",
+        "TB-R07"
+      ],
+      "scope": {
+        "allow": [
+          "package.json",
+          "package-lock.json",
+          "CHANGELOG.md"
+        ]
+      },
+      "description": "npm's @optitransfer/proofline is at 0.2.0-next.1 (latest and next, measured 2026-10-20), which predates PL-CRIT01. The version moves to 0.2.0-next.2 in package.json and package-lock.json (npm install --package-lock-only --ignore-scripts, empty user config), and the Unreleased heading in CHANGELOG.md becomes \"## 0.2.0-next.2 (2026-10-20)\". The owner publishes it from his terminal from the landed commit (npm publish --access public --tag next --auth-type=web) and records the registry's version and bin.",
+      "criteria": [
+        {
+          "id": "PL-NEXT2-AC01",
+          "mandatory": true,
+          "statement": "package.json says 0.2.0-next.2",
+          "verifier": {
+            "type": "grep",
+            "file": "package.json",
+            "pattern": "\"version\":\\s*\"0\\.2\\.0-next\\.2\"",
+            "must": "present"
+          }
+        },
+        {
+          "id": "PL-NEXT2-AC02",
+          "mandatory": true,
+          "statement": "the changelog dates 0.2.0-next.2 and keeps no Unreleased heading",
+          "verifier": {
+            "type": "command",
+            "cmd": "node -e \"const s=require('fs').readFileSync('CHANGELOG.md','utf8');process.exit(/^## 0\\.2\\.0-next\\.2 \\(2026-10-[0-9]{2}\\)$/m.test(s)&&!/^## Unreleased$/m.test(s)?0:1)\"",
+            "expect_exit": 0,
+            "timeout_s": 60
+          }
+        },
+        {
+          "id": "PL-NEXT2-AC03",
+          "mandatory": true,
+          "statement": "the packed tarball is 0.2.0-next.2, holds bin/proofline.mjs and the manifest names it as the proofline command",
+          "verifier": {
+            "type": "command",
+            "cmd": "sh -c \"npm pack --dry-run --json --ignore-scripts > /tmp/pl-pack.json && node -e \\\"const p=require('/tmp/pl-pack.json')[0];process.exit(p.version==='0.2.0-next.2'&&p.files.some(f=>f.path==='bin/proofline.mjs')&&require('./package.json').bin.proofline==='bin/proofline.mjs'?0:1)\\\"\"",
+            "expect_exit": 0,
+            "timeout_s": 300
+          }
+        },
+        {
+          "id": "PL-NEXT2-AC04",
+          "mandatory": true,
+          "statement": "Full suite exits 0, nothing hung",
+          "verifier": {
+            "type": "command",
+            "cmd": "npm test",
+            "expect_exit": 0,
+            "timeout_s": 1800
+          }
+        },
+        {
+          "id": "PL-NEXT2-AC05",
+          "mandatory": true,
+          "statement": "Changes stay in scope",
+          "verifier": {
+            "type": "scope"
+          }
+        },
+        {
+          "id": "PL-NEXT2-AC06",
+          "mandatory": true,
+          "statement": "Committed and pushed to main",
+          "verifier": {
+            "type": "pushed"
+          }
+        }
+      ],
+      "stop_rules": [
+        "If a criterion cannot pass without loosening a threshold, skipping or deleting a test, raising a timeout, widening the scope or editing the contract: STOP and report file:line and the measured value.",
+        "If the environment a criterion needs is missing (repository admin, npm owner login, Actions, a second machine): report BLOCKED naming the missing item; never mark it passed by hand."
+      ]
     }
   ],
-  "notes": "Each task carries the requirement ids it covers; next --owner hands out the tasks of one owner. Cross-repository needs are written into task text as an external dependency and reported BLOCKED when absent. PL-SELF stays: proofline verifies itself with its own contract. | 2026-10-17: 15 review tasks appended (COVERAGE.md maps every finding and open issue). Scope criterion added to PL-SELF, DG-PL so the contract passes its own validator. | 2026-10-17: PL-SETUP01 (setup pins the owner key) and PL-NEXT1 (0.2.0-next.1 carries the bin) appended.",
+  "notes": "Each task carries the requirement ids it covers; next --owner hands out the tasks of one owner. Cross-repository needs are written into task text as an external dependency and reported BLOCKED when absent. PL-SELF stays: proofline verifies itself with its own contract. | 2026-10-17: 15 review tasks appended (COVERAGE.md maps every finding and open issue). Scope criterion added to PL-SELF, DG-PL so the contract passes its own validator. | 2026-10-17: PL-SETUP01 (setup pins the owner key) and PL-NEXT1 (0.2.0-next.1 carries the bin) appended. | 2026-10-20: PL-CRIT01 (crit:<id> claims for tasks proved by a command; leading (?i)/(?s) grep flags) and PL-NEXT2 (0.2.0-next.2) appended.",
   "ci": {
     "tasks": [
       "PL-SELF"

proofline.json.sig

@@ -0,0 +1 @@
+{"payloadType":"application/vnd.proofline.contract-sha256","payload":"OWQzM2VjOWJlZWUxYmQyMmViNmJjZjk5NzVhZmRkN2UyZGJkZTc0MmVjMDQwYjE4NzdkMTE3YzMyMTBlZDMzNA==","signatures":[{"keyid":"kb71a7ce9e7b76859","sig":"ckpux3El8CYsS74U7YeISFDbO7cKboLT8U/Luj/4XXoKUXNLV3h0AOSTwPMKMi/B0HfuL4EGv2e61LD2AnGADA=="}]}

src/checks.mjs

@@ -212,7 +212,9 @@ export const verifiers = {
     const f = path.join(ctx.root, v.file);
     if (!fs.existsSync(f)) return { status: 'fail', detail: `${v.file} not found` };
     const txt = fs.readFileSync(f, 'utf8');
-    const re = new RegExp(v.pattern, v.flags || 'm');
+    // A leading (?i), (?s) or (?m) is how Python writes these flags; JavaScript refuses it as an invalid group.
+    const lead = String(v.pattern).match(/^\(\?([ims]+)\)/);
+    const re = new RegExp(lead ? v.pattern.slice(lead[0].length) : v.pattern, [...new Set((v.flags || 'm') + (lead ? lead[1] : ''))].join(''));
     const m = txt.match(re);
     const want = (v.must || 'present') === 'present';
     if (!!m === want) return { status: 'pass', detail: `/${v.pattern}/ ${want ? 'present' : 'absent'} in ${v.file}`, evidence: m ? { line: txt.slice(0, m.index).split('\n').length, text: m[0] } : undefined };

src/claims.mjs

@@ -4,6 +4,22 @@ import { git, onRemote } from './util.mjs';
 const DONE = /^(done|complete(d)?|verified|pass(ed)?|fixed|finished|green)$/i;
 const HEDGE = /\b(should work|likely|probably|untested|not tested|later|follow[- ]?up|out of scope|defer(red)?|todo|next session|in a future)\b/i;
 
+// `crit:<criterion id>` in the tests column names one of the task's own executed criteria instead of a file in the
+// repo, for a task whose proof is a command run by the gate and not a test it adds (an intake or measurement task).
+const CRIT = /^crit:(\S+)$/i;
+const EXECUTED = ['command', 'metric', 'test_red_green', 'docs'];
+
+/** Why `crit:<id>` does not prove the claim, or nothing: it must be a criterion of this task that ran a command in this gate and passed. */
+export function critIssues(id, task, res) {
+  const rows = res?.criteria;
+  if (!Array.isArray(rows)) return [`named criterion "crit:${id}" cannot be checked: the gate passed no criterion results`];
+  const r = rows.find((x) => x.id === id);
+  if (!r) return [`named criterion "crit:${id}" is not a criterion of ${task} that this gate ran`];
+  if (!EXECUTED.includes(r.type)) return [`named criterion "crit:${id}" is a ${r.type} check, which runs no test: name a command, metric, test_red_green or docs criterion`];
+  if (r.status !== 'pass') return [`named criterion "crit:${id}" is ${String(r.status).toUpperCase()} in this gate, not PASS`];
+  return [];
+}
+
 /** Lines: TASK-ID | status | commit | tests | measured vs target | missing */
 export function parseClaims(text, taskIds) {
   const out = [];
@@ -42,6 +58,8 @@ export function checkClaim(root, claim, res, { requirePushed = false, remote = '
   if (claimedDone && !names.length) issues.push('claimed done without naming the test(s) that prove it');
   for (const n of names) {
     const base = n.replace(/\s*\(.*\)$/, '').split(/\s+/)[0];
+    const crit = base.match(CRIT);
+    if (crit) { issues.push(...critIssues(crit[1], claim.id, res)); continue; }
     if (!git(['grep', '-q', '-F', base], root).ok && !git(['ls-files', '--', `*${base}*`], root).out) issues.push(`named test "${base}" not found in the repo`);
   }
   const keys = Object.keys(res.measured || {});

src/engine.mjs

@@ -193,7 +193,7 @@ export function verifyTask(root, taskId, opts = {}) {
   for (const a of t.criteria) if (a.verifier.type === 'metric') metricLabels[a.verifier.name || a.verifier.key] = a.verifier.labels || [];
   const claimResults = [];
   if (claims) {
-    const res = { decision, failing: [...blocked, ...failing], measured: ctx.measured, metricLabels };
+    const res = { decision, failing: [...blocked, ...failing], measured: ctx.measured, metricLabels, criteria: results.map(({ id, type, status }) => ({ id, type, status })) };
     const needPush = t.criteria.some((a) => a.verifier.type === 'pushed');
     for (const cl of parseClaims(claims, [t.id])) claimResults.push(checkClaim(root, cl, res, { requirePushed: needPush, ...pushTarget(c, t) }));
     for (const cr of claimResults) if (!cr.ok) failing.push(...cr.issues.map((i) => `claim mismatch: ${i}`));