Gladex Agent Logs

new_sandbox() {
  local base="${TMPDIR:-/tmp}/opencode"
  mkdir -p "$base" 2>/dev/null || base="/tmp"
  SB=$(mktemp -d "$base/qsc.XXXXXX") || { echo "cannot create sandbox" >&2; exit 1; }
  mkdir -p "$SB/repo"
}

cleanup() { [ -n "$SB" ] && rm -rf "$SB"; }
trap cleanup EXIT

if [ ! -x "$TOOL" ]; then
  echo "FAIL: tool not found/executable: $TOOL"
  exit 1
fi

# --- A: the healthy shape passes ----------------------------------------------
new_sandbox
healthy "$SB/repo"
run_tool "$SB/repo"
assert_rc "A1 healthy repo exits 0" 0 "$RC"
assert_contains "A2 human output says OK" "OK -" "$OUT"
assert_contains "A3 human output names the frozen count" "111 item line(s) frozen" "$OUT"
assert_contains "A4 human output names the newest version" "[0.4.67]" "$OUT"

run_tool "$SB/repo" --format json
assert_rc "A5 healthy json exits 0" 0 "$RC"
assert_jget "A6 json ok=true" "$OUT" 'd["ok"]' "True"
assert_jget "A7 json exit_code=0" "$OUT" 'd["exit_code"]' "0"
assert_jget "A8 json violations is empty" "$OUT" 'len(d["violations"])' "0"
assert_jget "A9 json error is null" "$OUT" 'd["error"]' "None"
assert_jget "A10 frozen_items_actual=111" "$OUT" 'd["rules"]["frozen_items_actual"]' "111"
assert_jget "A11 frozen_items_expected=111" "$OUT" 'd["rules"]["frozen_items_expected"]' "111"
$ cd /data/repo && sed -n '1588,1700p' tests/test_queue_source.sh
# --- Q: R8 - the newest CHANGELOG entry must be named in PROGRESS.md ----------
# The one rule that reads the two files as one story. R1-R4/R6-R7 judge
# CHANGELOG, R5 judges PROGRESS, and before this section each could be green
# while the two disagreed about the same run - measured on the live tree: the
# [0.4.160] entry landed with its own PROGRESS append staged out of the tree
# and never re-appended, `grep -c "0.4.160" agent-logs/PROGRESS.md` -> 0 for
# about 23 minutes, tool exit 0 with violations [] the whole time. Every
# fixture below therefore keeps the OTHER rules green on purpose, so a red can
# only be R8's (the isolation discipline section D states for the freeze).
new_sandbox
healthy "$SB/repo"
run_tool "$SB/repo" --format json
assert_rc "Q1 the healthy repo still exits 0 with R8 in force" 0 "$RC"
assert_jget "Q2 healthy repo: newest_version_in_progress=true" \
  "$OUT" 'd["rules"]["newest_version_in_progress"]' "True"
assert_jget "Q3 healthy repo: no violations" "$OUT" 'len(d["violations"])' "0"

# The defect itself: the newest token deleted from PROGRESS, nothing else moved.
new_sandbox
healthy "$SB/repo"
python3 - "$SB/repo/agent-logs/PROGRESS.md" <<'PY'
import sys
p = sys.argv[1]
t = open(p, encoding="utf-8").read()
assert t.count("[0.4.67]") == 1, "Q fixture precondition: %d token(s)" % t.count("[0.4.67]")
open(p, "w", encoding="utf-8").write(t.replace("[0.4.67]", "the run", 1))
PY
FIX="$SB/repo"
run_tool "$FIX" --format json
assert_rc "Q4 a PROGRESS that never names the newest entry exits 1" 1 "$RC"
assert_jget "Q5 newest_version_in_progress=false there" \
  "$OUT" 'd["rules"]["newest_version_in_progress"]' "False"
assert_jget "Q6 exactly one violation - R8 alone, every other rule green" \
  "$OUT" 'len(d["violations"])' "1"
assert_jget "Q7 the freeze is untouched (111) - R1 cannot see this defect" \
  "$OUT" 'd["rules"]["frozen_items_actual"]' "111"
assert_jget "Q8 the pointer rule is still green - R4 cannot see it either" \
  "$OUT" 'd["rules"]["newest_points_at_progress"]' "True"
assert_jget "Q9 R5 still counts its bullet - the file is healthy, only behind" \
  "$OUT" 'd["rules"]["progress_bullets"]' "1"
run_tool "$FIX"
assert_contains "Q10 the human channel names the entry and the file it is missing from" \
  "[0.4.67] is not named in agent-logs/PROGRESS.md" "$OUT"

# A version IS named - just not the newest one. R8 judges the newest entry, so
# any older token is the wrong claim.
new_sandbox
healthy "$SB/repo"
python3 - "$SB/repo/agent-logs/PROGRESS.md" <<'PY'
import sys
p = sys.argv[1]
t = open(p, encoding="utf-8").read()
open(p, "w", encoding="utf-8").write(t.replace("[0.4.67]", "[0.4.66]", 1))
PY
FIX="$SB/repo"
run_tool "$FIX" --format json
assert_rc "Q11 naming an OLDER version does not satisfy R8 (exit 1)" 1 "$RC"
assert_jget "Q12 the newest entry is still 0.4.67 - the rule reads the right one" \
  "$OUT" 'd["rules"]["newest_version"]' "0.4.67"
assert_jget "Q13 and it is still reported false" \
  "$OUT" 'd["rules"]["newest_version_in_progress"]' "False"
assert_jget "Q14 still exactly one violation" "$OUT" 'len(d["violations"])' "1"

# A version QUOTED in a fenced block is not a run recording itself - the same
# fence rule count_queue_bullets() applies, and the reason R8 shares it.
new_sandbox
healthy "$SB/repo"
python3 - "$SB/repo/agent-logs/PROGRESS.md" <<'PY'
import sys
open(sys.argv[1], "w", encoding="utf-8").write(
    "# PROGRESS\n"
    "\n"
    "## 2026-09-26 run\n"
    "- **Next-candidate queued, not actioned**: (1) still an entry, so R5 is green.\n"
    "```text\n"
    "a paste that says [0.4.67] while the run itself wrote nothing.\n"
    "```\n")
PY
FIX="$SB/repo"
run_tool "$FIX" --format json
assert_rc "Q15 a fence-only mention does not satisfy R8 (exit 1)" 1 "$RC"
assert_jget "Q16 newest_version_in_progress=false outside the fence" \
  "$OUT" 'd["rules"]["newest_version_in_progress"]' "False"
assert_jget "Q17 exactly one violation - the fence kept R5 green too" \
  "$OUT" 'len(d["violations"])' "1"

# The boundary, in both directions: `0.4.67` must not be answered by a longer
# token, nor found inside a longer one.
new_sandbox
healthy "$SB/repo"
python3 - "$SB/repo/agent-logs/PROGRESS.md" <<'PY'
import sys
p = sys.argv[1]
t = open(p, encoding="utf-8").read()
open(p, "w", encoding="utf-8").write(t.replace("[0.4.67]", "[0.4.670]", 1))
PY
run_tool "$SB/repo" --format json
assert_rc "Q18 the lookahead holds: 0.4.67 is not satisfied by 0.4.670" 1 "$RC"
assert_jget "Q19 and it is reported false, not merely red" \
  "$OUT" 'd["rules"]["newest_version_in_progress"]' "False"

new_sandbox
healthy "$SB/repo"
python3 - "$SB/repo/agent-logs/PROGRESS.md" <<'PY'
import sys
p = sys.argv[1]
t = open(p, encoding="utf-8").read()
open(p, "w", encoding="utf-8").write(t.replace("[0.4.67]", "[10.4.67]", 1))
PY
run_tool "$SB/repo" --format json
assert_rc "Q20 the lookbehind holds: 0.4.67 is not found inside 10.4.67" 1 "$RC"
assert_jget "Q21 and it is reported false" \
  "$OUT" 'd["rules"]["newest_version_in_progress"]' "False"
> build · mimo-v2.6-flash-free
Root cause identified. The lint docstring's own SCOPE rule says fixture tokens belong in a *foreign* series (`[1.0.0]`, `[9.9.9]` are "not citations of this file") — `0.4.670` sits in the live `0.4` series with no heading, so it's judged. Let me see how REGISTRY documents Q18:
$ cd /data/repo && grep -n "Q18\|Q19\|0\.4\.670\|lookahead" tools/REGISTRY.md | head; echo "=== context ==="; grep -n "Q (24)\|Q17\|Q20" tools/REGISTRY.md | head
3003:  difference); the boundary **both ways** (`0.4.670` and `10.4.67` neither
=== context ===
2991:- **Q (24)** `[0.4.161]`'s rule, one fixture per way the two files can disagree
$ cd /data/repo && sed -n '/^## test_queue_source.sh/,/^\*\*Tests/p' tools/REGISTRY.md | grep -n "Q\b\|Q (" | head; echo "--- find Q bullet ---"; awk '/^## test_queue_source.sh/{f=1} f&&/- \*\*Q \(/{p=1} p{print NR": "$0} p&&/^- \*\*[A-Z]/&&NR>1&&!/- \*\*Q \(/{if(++c>1)exit}' tools/REGISTRY.md | head -40
26:applied to the second file. The cases that DO exercise R8 are section **Q**'s
31:M 26, O 12, P 14, Q 24** (sums to 209).
124:- **Q (24)** `[0.4.161]`'s rule, one fixture per way the two files can disagree
187:**Status**: Green ✅ 2026-10-01 (**209/209**; `[0.4.161]` added section **Q**
--- find Q bullet ---
2991: - **Q (24)** `[0.4.161]`'s rule, one fixture per way the two files can disagree
2992:   about the same run, each holding **every other rule green** so a red can only
2993:   be R8's: the healthy repo (`newest_version_in_progress = true`, zero
2994:   violations); the defect itself — the newest token deleted from `PROGRESS.md`,
2995:   which is what `[0.4.160]` shipped while `grep -c` read **0** for ≈23 minutes —
2996:   → exit 1 with **exactly one** violation while the freeze still reads **111**,
2997:   the pointer rule still reads `true` and `progress_bullets` still counts its
2998:   entry (the file is *healthy*, only *behind*), plus the human channel naming
2999:   the entry and the file; an **older** version named instead (`[0.4.66]`) →
3000:   still false, and `newest_version` asserted to still be the one being judged;
3001:   a version quoted **inside a fence** → false (the `count_queue_bullets()`
3002:   fence rule, shared deliberately, with R5 green so the fence is the only
3003:   difference); the boundary **both ways** (`0.4.670` and `10.4.67` neither
3004:   satisfy nor hide the token) → false; and the **bare** spelling `0.4.67` →
3005:   exit 0, because R8 claims the version is *named*, not punctuated a particular
3006:   way — a bracket requirement would be a formatting rule wearing a defect's
3007:   costume.
3008: - **J (7) + M (26)** the mutation harness and its thirteen mutants.
3009: 
3010: **Mutations (13, copies only — the tool's md5 is unchanged by all of them)**:
3011: M1 drop the freeze equality; M2 drop the second-list rule; M3 drop the pointer
3012: rule; M4 drop the authoritative-list rule; M5 turn exit 3 into a pass; M6 break
3013: `### Queue` detection; M7 truncate `violations` to its first row; M8 drop fence
3014: handling in `parse()`; M9 put R5 back to `pg_text.count(...)`; M10 drop the
3015: fence check inside `count_queue_bullets()`; **M11 drop R6** (the duplicate
3016: filter `n > 1` → `n > 99`, so the `[0.4.102]` fixture exits **0** — the mutant
3017: reproduces the shipped defect rather than a symptom of it); **M12 make `parse()`
3018: skip a `### Queue` above the first version heading** (R7's tripwire: mutant
3019: exits **1** naming *"a heading was dropped"* and with **one** violation, while
3020: the real tool exits **0** on the same fixture — and the same mutant stays
3021: green on a healthy repo, so the red is attributable to the drop alone); and
3022: **M13 drop R8** (the `if not named:` arm turned off, so the stale-queue fixture
3023: — newest token deleted from `PROGRESS.md`, nothing else moved — exits **0**,
3024: i.e. exactly the state `[0.4.160]` carried for 23 minutes: tool green,
3025: `violations []`, the two files disagreeing; the real tool exits **1** with that
3026: violation as its only one).
3027: **M7 is the one that matters**: the
3028: mutant still exits **1**, identical to the truth for any exit-code check, so
3029: only `len(violations)` sees it — which is the reason the tool collects and
3030: reports every rule instead of short-circuiting. `J0` pins the harness itself: a
> build · mimo-v2.6-flash-free

Select another run log from the list above. Only files matching run-YYYYMMDD-HHMMSS-N.log are readable.

[Fri Oct  2 02:13:36 2026] 127.0.0.1:56246 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56246 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56256 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56256 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56264 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56264 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56280 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56280 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56292 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56292 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56300 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56300 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56308 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56308 Closing
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56314 Accepted
[Fri Oct  2 02:13:36 2026] 127.0.0.1:56314 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42972 Accepted
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42972 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42980 Accepted
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42980 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42984 Accepted
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42984 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42992 Accepted
[Fri Oct  2 02:14:18 2026] 127.0.0.1:42992 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:43006 Accepted
[Fri Oct  2 02:14:18 2026] 127.0.0.1:43006 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:43008 Accepted
[Fri Oct  2 02:14:18 2026] 127.0.0.1:43008 Closing
[Fri Oct  2 02:14:18 2026] 127.0.0.1:43020 Accepted
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43020 Closing
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43026 Accepted
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43026 Closing
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43034 Accepted
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43034 Closing
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43042 Accepted
[Fri Oct  2 02:14:19 2026] 127.0.0.1:43042 Closing
[Fri Oct  2 02:15:16 2026] 127.0.0.1:42406 Accepted
[Fri Oct  2 02:15:16 2026] 127.0.0.1:42406 Closing
[Fri Oct  2 02:15:16 2026] 127.0.0.1:42414 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42414 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42428 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42428 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42444 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42444 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42458 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42458 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42474 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42474 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42490 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42490 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42502 Accepted
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42502 Closing
[Fri Oct  2 02:15:17 2026] 127.0.0.1:42518 Accepted
[Fri Oct  2 02:15:18 2026] 127.0.0.1:42518 Closing
[Fri Oct  2 02:15:18 2026] 127.0.0.1:42532 Accepted
[Fri Oct  2 02:15:18 2026] 127.0.0.1:42532 Closing
[Fri Oct  2 02:15:59 2026] 127.0.0.1:52524 Accepted
[Fri Oct  2 02:15:59 2026] 127.0.0.1:52524 Closing
[Fri Oct  2 02:20:40 2026] 127.0.0.1:51030 Accepted