The PR-verdict gate (.github/workflows/pr-verdict.yml) validates the opening line of a PR body with:
grep -qE "^#{0,6}[[:space:]]*[^[:alnum:]]*[[:space:]]*SAFE TO MERGE([[:space:]]|$)"[^[:alnum:]]* means "zero or more non-alphanumeric characters" — it does not require the specific 🟢 emoji, or any emoji at all. All three of these match as a valid safe verdict:
## SAFE TO MERGE
## 🔴 SAFE TO MERGE
## !!! SAFE TO MERGE
The color/marker is decorative to the regex; only the words govern.
Not a regression
Verified this is pre-existing, not introduced by #202: running the same three bare-marker lines through the pre-#202 version of the regex (...SAFE TO MERGE[[:space:]]*$, no trailing-clause allowance) matches all three identically. The looseness has been present since the gate's original version (#110).
Open question
Is text-governs the intended design (the emoji is a human-readable affordance, the words are the contract), or should the specific marker codepoint be part of what the regex requires? Both are defensible; this issue is to make the choice deliberate rather than leaving it as an accident of the character class chosen in #110.
The PR-verdict gate (
.github/workflows/pr-verdict.yml) validates the opening line of a PR body with:grep -qE "^#{0,6}[[:space:]]*[^[:alnum:]]*[[:space:]]*SAFE TO MERGE([[:space:]]|$)"[^[:alnum:]]*means "zero or more non-alphanumeric characters" — it does not require the specific 🟢 emoji, or any emoji at all. All three of these match as a validsafeverdict:The color/marker is decorative to the regex; only the words govern.
Not a regression
Verified this is pre-existing, not introduced by #202: running the same three bare-marker lines through the pre-#202 version of the regex (
...SAFE TO MERGE[[:space:]]*$, no trailing-clause allowance) matches all three identically. The looseness has been present since the gate's original version (#110).Open question
Is text-governs the intended design (the emoji is a human-readable affordance, the words are the contract), or should the specific marker codepoint be part of what the regex requires? Both are defensible; this issue is to make the choice deliberate rather than leaving it as an accident of the character class chosen in #110.