Harden prompted Hermes session handoff
Bind option-looking prompts to one-shot mode, require a successful completed usage report before resuming, replay the prompt after first-run setup, and reject Hermes runtimes that lack the session-report capability. Co-Authored-By: Codex XHigh <noreply@openai.com>
This commit is contained in:
Spencer BullandCodex XHigh committed
1 parent
288e387a22
commit
36353296aa
4 files changed
+138
-20
No files matched your search
@@ -11,12 +11,21 @@ mock_bin="$test_tmp/bin"
|
||||
oneshot_log="$test_tmp/oneshot"
|
||||
resume_log="$test_tmp/resume"
|
||||
source_log="$test_tmp/source"
|
||||
setup_log="$test_tmp/setup"
|
||||
setup_marker="$test_tmp/setup-complete"
|
||||
mkdir -p "$mock_bin"
|
||||
|
||||
cat >"$mock_bin/hermes" <<'SH'
|
||||
#!/bin/bash
|
||||
|
||||
if [[ " $* " == *" --oneshot "* ]]; then
|
||||
if [[ ${1:-} == "setup" ]]; then
|
||||
printf '%s\0' "$@" >"$HERMES_TEST_SETUP_LOG"
|
||||
[[ ${HERMES_TEST_SETUP_FAIL:-false} == "false" ]] || exit 43
|
||||
touch "$HERMES_TEST_SETUP_MARKER"
|
||||
exit
|
||||
fi
|
||||
|
||||
if [[ " $* " == *" --oneshot="* ]]; then
|
||||
printf '%s\0' "$@" >"$HERMES_TEST_ONESHOT_LOG"
|
||||
printf '%s' "${HERMES_SESSION_SOURCE:-}" >"$HERMES_TEST_SOURCE_LOG"
|
||||
|
||||
@@ -28,8 +37,19 @@ if [[ " $* " == *" --oneshot "* ]]; then
|
||||
shift
|
||||
done
|
||||
|
||||
if [[ ${HERMES_TEST_NEEDS_SETUP:-false} == "true" && ! -e $HERMES_TEST_SETUP_MARKER ]]; then
|
||||
printf '{"session_id":null,"completed":null,"failed":true,"failure":"No inference provider configured. Run hermes model."}\n' >"$usage"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
[[ ${HERMES_TEST_ONESHOT_FAIL:-false} == "false" ]] || exit 42
|
||||
[[ ${HERMES_TEST_USAGE_FAIL:-false} == "false" ]] && printf '{"session_id":"session-123"}\n' >"$usage"
|
||||
if [[ ${HERMES_TEST_USAGE_FAIL:-false} == "false" ]]; then
|
||||
completed=true
|
||||
failed=false
|
||||
[[ ${HERMES_TEST_USAGE_INCOMPLETE:-false} == "false" ]] || completed=false
|
||||
[[ ${HERMES_TEST_USAGE_FAILED:-false} == "false" ]] || failed=true
|
||||
printf '{"session_id":"session-123","completed":%s,"failed":%s}\n' "$completed" "$failed" >"$usage"
|
||||
fi
|
||||
printf '%s\n' response
|
||||
exit
|
||||
fi
|
||||
@@ -43,18 +63,19 @@ export PATH="$mock_bin:$PATH"
|
||||
export HERMES_TEST_ONESHOT_LOG="$oneshot_log"
|
||||
export HERMES_TEST_RESUME_LOG="$resume_log"
|
||||
export HERMES_TEST_SOURCE_LOG="$source_log"
|
||||
export HERMES_TEST_SETUP_LOG="$setup_log"
|
||||
export HERMES_TEST_SETUP_MARKER="$setup_marker"
|
||||
|
||||
sentinel="$test_tmp/hermes-seed-must-stay-literal"
|
||||
prompt="!Crash /quit {!touch $sentinel}"$'\ntrailing\\'
|
||||
prompt="--help !Crash /quit {!touch $sentinel}"$'\ntrailing\\'
|
||||
"$ROOT/bin/omarchy-agent-hermes" "$prompt" >/dev/null
|
||||
|
||||
mapfile -d '' -t oneshot_args <"$oneshot_log"
|
||||
(( ${#oneshot_args[@]} == 5 )) || fail "Hermes literal seed has five one-shot arguments"
|
||||
(( ${#oneshot_args[@]} == 4 )) || fail "Hermes literal seed has four one-shot arguments"
|
||||
[[ ${oneshot_args[0]} == "--yolo" ]] || fail "Hermes literal seed enables yolo mode"
|
||||
[[ ${oneshot_args[1]} == "--usage-file" ]] || fail "Hermes literal seed requests the session report"
|
||||
usage_file=${oneshot_args[2]}
|
||||
[[ ${oneshot_args[3]} == "--oneshot" && ${oneshot_args[4]} == "$prompt" ]] ||
|
||||
fail "Hermes literal seed remains one argument"
|
||||
[[ ${oneshot_args[3]} == "--oneshot=$prompt" ]] || fail "Hermes literal seed binds option-looking prompts as data"
|
||||
[[ ! -e $usage_file ]] || fail "Hermes literal seed removes its session report"
|
||||
[[ ! -e $sentinel ]] || fail "Hermes literal seed never executes prompt interpolation"
|
||||
[[ ! -s $source_log ]] || fail "Hermes literal seed preserves native CLI session metadata"
|
||||
@@ -77,3 +98,29 @@ if HERMES_TEST_USAGE_FAIL=true "$ROOT/bin/omarchy-agent-hermes" missing-session
|
||||
fi
|
||||
[[ ! -s $resume_log ]] || fail "Hermes literal seed does not guess which session to resume"
|
||||
pass "Hermes resumes only the session recorded by the initial turn"
|
||||
|
||||
for state in INCOMPLETE FAILED; do
|
||||
: >"$resume_log"
|
||||
if env "HERMES_TEST_USAGE_$state=true" "$ROOT/bin/omarchy-agent-hermes" "${state,,}" >/dev/null 2>&1; then
|
||||
fail "Hermes literal seed rejects a reported ${state,,} initial turn"
|
||||
fi
|
||||
[[ ! -s $resume_log ]] || fail "Hermes literal seed does not resume a reported ${state,,} initial turn"
|
||||
done
|
||||
pass "Hermes resumes only completed successful initial turns"
|
||||
|
||||
: >"$resume_log"
|
||||
HERMES_TEST_NEEDS_SETUP=true "$ROOT/bin/omarchy-agent-hermes" setup-first >/dev/null
|
||||
mapfile -d '' -t setup_args <"$setup_log"
|
||||
[[ ${setup_args[*]} == "setup" ]] || fail "Hermes runs setup when no inference provider is configured"
|
||||
mapfile -d '' -t resume_args <"$resume_log"
|
||||
[[ ${resume_args[*]} == "chat --yolo --tui --resume session-123" ]] ||
|
||||
fail "Hermes replays the prompted turn after setup and resumes it"
|
||||
pass "Hermes completes first-run setup before replaying the prompt"
|
||||
|
||||
rm -f "$setup_marker"
|
||||
: >"$resume_log"
|
||||
if HERMES_TEST_NEEDS_SETUP=true HERMES_TEST_SETUP_FAIL=true "$ROOT/bin/omarchy-agent-hermes" setup-cancelled >/dev/null 2>&1; then
|
||||
fail "Hermes reports a failed first-run setup"
|
||||
fi
|
||||
[[ ! -s $resume_log ]] || fail "Hermes does not resume when first-run setup fails"
|
||||
pass "Hermes stops when first-run setup does not complete"
|
||||
@@ -26,7 +26,10 @@ SH
|
||||
cat >"$mock_bin/mise" <<'SH'
|
||||
#!/bin/bash
|
||||
printf '%s\0' "$@" >>"$OMARCHY_TEST_MISE_LOG"
|
||||
[[ $1 == "where" && ${OMARCHY_TEST_MISE_WHERE_OK:-0} == 1 ]] && exit 0
|
||||
if [[ $1 == "where" && ${OMARCHY_TEST_MISE_WHERE_OK:-0} == 1 ]]; then
|
||||
printf '%s\n' "$OMARCHY_TEST_MISE_ROOT"
|
||||
exit 0
|
||||
fi
|
||||
[[ $1 != "where" ]]
|
||||
SH
|
||||
|
||||
@@ -35,6 +38,7 @@ chmod +x "$mock_bin"/*
|
||||
run_installer() {
|
||||
OMARCHY_TEST_DESKTOP_INSTALLED="$1" \
|
||||
OMARCHY_TEST_MISE_WHERE_OK="${OMARCHY_TEST_MISE_WHERE_OK:-0}" \
|
||||
OMARCHY_TEST_MISE_ROOT="$test_tmp/mise" \
|
||||
OMARCHY_TEST_MISE_LOG="$mise_log" \
|
||||
HOME="$test_home" \
|
||||
PATH="$mock_bin:$PATH" \
|
||||
@@ -88,10 +92,17 @@ pass "takeover removes an unhealthy mise copy"
|
||||
rm -rf "$test_home/.hermes"
|
||||
rm -f "$test_home/.local/bin/hermes"
|
||||
run_installer 1 --check && fail "--check reports Hermes missing before the app installs it"
|
||||
# The venv command answers --version, as the real one does: foreign wrappers
|
||||
# below exec it, and the installer probes them by running exactly that.
|
||||
# The venv command answers the readiness probes, as the real one does: foreign
|
||||
# wrappers below exec it, and the installer runs both before trusting them.
|
||||
mkdir -p "$test_home/.hermes/hermes-agent/venv/bin"
|
||||
printf '%s\n' "#!/bin/bash" 'echo "hermes-agent 0.0.0-test"' >"$test_home/.hermes/hermes-agent/venv/bin/hermes"
|
||||
cat >"$test_home/.hermes/hermes-agent/venv/bin/hermes" <<'SH'
|
||||
#!/bin/bash
|
||||
if [[ ${1:-} == "--help" ]]; then
|
||||
[[ ${OMARCHY_TEST_HERMES_CAPABLE:-1} == 1 ]] && echo "--usage-file PATH"
|
||||
else
|
||||
echo "hermes-agent 0.0.0-test"
|
||||
fi
|
||||
SH
|
||||
chmod +x "$test_home/.hermes/hermes-agent/venv/bin/hermes"
|
||||
run_installer 1 --check && fail "--check waits for the install to finish, not just the venv"
|
||||
touch "$test_home/.hermes/hermes-agent/.hermes-bootstrap-complete"
|
||||
@@ -123,6 +134,14 @@ run_installer 0 --now || fail "--now over a foreign hermes command returns succe
|
||||
fail "a foreign hermes command is left untouched"
|
||||
pass "a foreign hermes command is preserved and satisfies --check"
|
||||
|
||||
OMARCHY_TEST_HERMES_CAPABLE=0 run_installer 0 --check &&
|
||||
fail "--check rejects a foreign Hermes without prompted-session reports"
|
||||
OMARCHY_TEST_HERMES_CAPABLE=0 run_installer 0 &&
|
||||
fail "installing refuses a foreign Hermes without prompted-session reports"
|
||||
[[ $(cat "$test_home/.local/bin/hermes") == "$official_body" ]] ||
|
||||
fail "an older foreign Hermes command is left untouched"
|
||||
pass "a foreign Hermes must support prompted-session reports"
|
||||
|
||||
# Broken foreign paths are still foreign. They cannot be used, so --check says
|
||||
# so and the installer refuses rather than replacing them.
|
||||
printf '%s\n' "$official_body" >"$test_home/.local/bin/hermes"
|
||||
@@ -211,6 +230,13 @@ grep -q "stale template" "$test_home/.local/bin/hermes" && fail "reinstalling re
|
||||
grep -q "exec env -u UV_PYTHON mise x" "$test_home/.local/bin/hermes" || fail "the refreshed stub is the current template"
|
||||
pass "reinstalling refreshes the Omarchy stub"
|
||||
|
||||
mkdir -p "$test_tmp/mise/hermes-agent/lib/python$python_pin"
|
||||
: >"$mise_log"
|
||||
OMARCHY_TEST_MISE_WHERE_OK=1 run_installer 0 || fail "reinstalling replaces an older owned Hermes environment"
|
||||
tr '\0' '\n' <"$mise_log" | grep -q '^rm$' || fail "an older owned Hermes environment is removed from mise config"
|
||||
tr '\0' '\n' <"$mise_log" | grep -q '^uninstall$' || fail "an older owned Hermes environment is uninstalled"
|
||||
pass "reinstalling replaces an older owned Hermes environment"
|
||||
|
||||
# install/user/mise.sh is sourced by install/user/all.sh through run_logged,
|
||||
# which runs it under `bash -eE` and hands its exit code back to
|
||||
# omarchy-provision-user's `set -euo pipefail`. Everything that finalizes a user
|
||||
@@ -349,8 +375,14 @@ run_ready_check() {
|
||||
|
||||
run_ready_check && fail "--check rejects the app's wrapper when its runtime is gone"
|
||||
|
||||
printf '%s\n' "#!/bin/bash" 'echo "hermes-agent 0.0.0-test"' \
|
||||
>"$ready_home/.hermes/hermes-agent/venv/bin/hermes"
|
||||
cat >"$ready_home/.hermes/hermes-agent/venv/bin/hermes" <<'SH'
|
||||
#!/bin/bash
|
||||
if [[ ${1:-} == "--help" ]]; then
|
||||
echo "--usage-file PATH"
|
||||
else
|
||||
echo "hermes-agent 0.0.0-test"
|
||||
fi
|
||||
SH
|
||||
chmod +x "$ready_home/.hermes/hermes-agent/venv/bin/hermes"
|
||||
run_ready_check || fail "--check accepts the app's wrapper once it runs"
|
||||
pass "readiness runs the app's command rather than trusting its marker"
|
||||
Reference in new issue
Block a user