Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
Rust
fix(hooks): JSON escaping was a no-op on BusyBox awk (#733)
Abhishek Sharmac83076b3a1d8
diff --git a/hooks/_lib.sh b/hooks/_lib.shindex d4f2fa23..a3080670 100644--- a/hooks/_lib.sh+++ b/hooks/_lib.sh@@ -516,12 +516,32 @@ ai_memory_get_handoff() {# handoff in hookSpecificOutput.additionalContext).ai_memory_json_string() {awk '- BEGIN { printf "\"" }+ # BusyBox awk applies backslash processing to a gsub REPLACEMENT+ # string; gawk, mawk and one-true-awk pass it through literally. Every+ # replacement below carries a backslash, so on BusyBox all four escapes+ # were silently no-ops (#733): a backslash stayed bare, a quote stayed+ # bare, and -- not in the report, but the same root cause -- \t and \r+ # collapsed to the letters "t" and "r", corrupting content rather than+ # only breaking the framing.+ #+ # Detect the behaviour once instead of guessing at it, and pre-double+ # the replacements where they will be halved. Done with gsub rather+ # than a split/concat loop on purpose: concatenating per occurrence is+ # quadratic in the value length, which is the cost #727 was about.+ BEGIN {+ probe = "X"; gsub(/X/, "\\\\", probe)+ if (length(probe) == 1) {+ BS = "\\\\\\\\"; QT = "\\\\\""; TB = "\\\\t"; CR = "\\\\r"+ } else {+ BS = "\\\\"; QT = "\\\""; TB = "\\t"; CR = "\\r"+ }+ printf "\""+ }{- gsub(/\\/, "\\\\")- gsub(/"/, "\\\"")- gsub(/\t/, "\\t")- gsub(/\r/, "\\r")+ gsub(/\\/, BS)+ gsub(/"/, QT)+ gsub(/\t/, TB)+ gsub(/\r/, CR)printf "%s%s", sep, $0sep = "\\n"}