Connects Jira to AI agents for seamless issue tracking management.

Install

mkdir -p .claude/skills/integration-jira && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/16251" && unzip -o skill.zip -d .claude/skills/integration-jira && rm skill.zip

Installs to .claude/skills/integration-jira

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Connect Jira (Atlassian Cloud) to a self-hosted Hermes Agent over SSH so the agent can search, create, and update issues. Wires Atlassian's official remote MCP server (Rovo) with a static API token. Idempotent and rollback-safe. Works from Claude Code, Codex, Cursor, Hermes itself, and Gemini CLI.
298 charsno explicit “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
Advanced

Key capabilities

  • →Connect Jira (Atlassian Cloud) to a self-hosted Hermes Agent over SSH
  • →Register the Hermes Agent with Atlassian's remote MCP server (Rovo)
  • →Store Atlassian API tokens securely on the VPS
  • →Reload the Hermes gateway for configuration changes
  • →Verify Jira integration through logs and smoke tests

How it works

The skill connects Jira to a Hermes agent on a VPS via SSH by securely storing the API token, registering the agent with Atlassian's MCP server, and reloading the gateway. It includes idempotency checks and dry runs.

Inputs & outputs

You give it
VPS IP/hostname, SSH user, Atlassian email, Jira API token, confirmation of org-admin toggle for Rovo MCP
You get back
Jira (Atlassian Cloud) connected to Hermes Agent, verified MCP registration

When to use integration-jira

  • →Creating Jira tickets
  • →Searching issue backlogs
  • →Updating ticket statuses

About this skill

/integration-jira — connect jira to Hermes over SSH

Compatibility: source-verified for Hermes 0.23.2, commit ae2a81ccc00e9b7c06301523e5bb09608397149c. Host-installed default profile only. Vendor live validation is pending; source and sandbox tests do not prove a live connection.

Before you start — gather (ask once, in one batch)

VariableWhatWhere to get it
$VPS_IPIP/hostname of the VPS running HermesUser's hosting dashboard
$VPS_USERSSH user (typically root)User's hosting dashboard
$ATLASSIAN_EMAILEmail of the Atlassian account the agent acts asThe user's Atlassian account
$JIRA_TOKENAtlassian API token (opaque string)https://id.atlassian.com/manage-profile/security/api-tokens → Create API token

The MCP server runs within that user's Jira permissions — pick the right account.

Also provide $JIRA_SITE=https://your-team.atlassian.net; this validates your Jira identity before configuring the native Atlassian OAuth MCP.

Use one shell session. DRY_RUN=1 runs the read-only gates and prints the plan, then exits before snapshot or configuration writes. FORCE=1 permits replacing existing MCP configuration after snapshot; existing native skill directories are always preserved.

Step 0 — shared remote, lifecycle and recovery helpers

# Run every block of this skill in ONE shell session: later steps call these functions.
# If your harness starts a fresh shell per command, re-run this block first.
SKILL_NAME=integration-jira
# Non-login ssh shells miss ~/.local/bin, where the Hermes installer puts `hermes`.
remote() { ssh -o BatchMode=yes "$VPS_USER@$VPS_IP" "export PATH=\"\$HOME/.local/bin:\$PATH\"; $*"; }
# gw_ready [SECONDS]: native readiness. Succeeds only when `gateway ensure --json` exits 0 and reports "state":"ready"
# (the transient pending/draining codes 5 and 6 are retried twice). It never greps `gateway status` text, because the stopped
# gateway prints "✗ Gateway is not running", which contains "running".
gw_ready() {
  remote "T=${1:-60} bash -s" <<'REMOTE'
n=0
while :; do
  out=$(hermes -p default gateway ensure --json --timeout "$T") && rc=0 || rc=$?
  [ "$rc" = 0 ] && break
  n=$((n + 1))
  { [ "$rc" = 5 ] || [ "$rc" = 6 ]; } && [ "$n" -lt 3 ] || { printf '%s\n' "$out" >&2; exit 1; }
  sleep 3
done
case $out in *'"state":"ready"'*) printf '%s\n' "$out" ;; *) printf '%s\n' "$out" >&2; exit 1 ;; esac
REMOTE
}
# gw_reload: restart the default-profile gateway natively and require a ready gateway with a NEW instance_id.
# `--detach` keeps `restart` from running the gateway inside the ssh shell when no service is installed. With live
# sessions and no terminal, restart refuses; GW_INTERRUPT=1 adds -y and interrupts them deliberately (never set by default).
# Side effect: the pre-restart `ensure` read that captures the old instance_id STARTS a gateway when none is running (ensure never
# installs a service, but it does start one). A reload implies the gateway should run, so that is accepted; read-only skills never call gw_reload.
gw_reload() {
  local old new out interrupt
  interrupt=""
  if [ "${GW_INTERRUPT:-0}" = 1 ]; then interrupt=1; fi
  # Keep an endpoint identity even when ensure exits 6 (draining/restart_wait).
  out=$(remote 'hermes -p default gateway ensure --json --timeout 10 2>/dev/null') || :
  old=$(printf '%s\n' "$out" | sed -n 's/.*"instance_id":"\([^"]*\)".*/\1/p')
  remote "hermes -p default gateway restart --detach${interrupt:+ -y}" \
    || { echo "gw_reload: gateway restart failed or was refused (live sessions? re-run with GW_INTERRUPT=1 only if interrupting them is acceptable)." >&2; return 1; }
  out=$(gw_ready) || return 1
  new=$(printf '%s' "$out" | sed -n 's/.*"instance_id":"\([^"]*\)".*/\1/p')
  if [ -n "$old" ] && [ "$new" = "$old" ]; then echo "gw_reload: the gateway instance did not change, so it was not restarted." >&2; return 1; fi
  printf '%s\n' "$out"
}

# snapshot: copy ~/.hermes/.env and config.yaml aside BEFORE the first write.
# The snapshot dir doubles as the "run in progress" marker that rollback looks for.
snapshot() {
  remote "SKILL_NAME='$SKILL_NAME' bash -s" <<'REMOTE'
set -e
p="$HOME/.hermes/.hstack-backup"
d="$p/$SKILL_NAME"
# The active recovery point is published only after every artifact has been verified.
if [ -e "$d" ] || [ -L "$d" ]; then
  echo "snapshot: $d exists — a previous $SKILL_NAME run did not finish. Run rollback first (restores the state before that run), then re-run." >&2
  exit 1
fi
mkdir -p "$p" && chmod 700 "$p"
lock="$d.creating"
mkdir "$lock" 2>/dev/null || { echo "snapshot: another $SKILL_NAME snapshot is being created; retry after it finishes." >&2; exit 1; }
t=""
trap '[ -z "$t" ] || rm -rf "$t"; rmdir "$lock" 2>/dev/null || true' EXIT
[ ! -e "$d" ] && [ ! -L "$d" ] || { echo "snapshot: a recovery point appeared; nothing was written." >&2; exit 1; }
t=$(mktemp -d "$p/.snapshot-$SKILL_NAME.XXXXXX")
for f in .env config.yaml; do
  if [ -f "$HOME/.hermes/$f" ]; then
    cp -p "$HOME/.hermes/$f" "$t/$f"
    cmp -s "$HOME/.hermes/$f" "$t/$f"
    chmod 600 "$t/$f"
  elif [ -e "$HOME/.hermes/$f" ] || [ -L "$HOME/.hermes/$f" ]; then
    echo "snapshot: $f is not a readable regular file; nothing was written." >&2; exit 1
  else
    : > "$t/$f.absent"
    chmod 600 "$t/$f.absent"
  fi
done

for f in .env config.yaml; do
  for entry in "$f" "$f.absent"; do
    if [ -f "$t/$entry" ]; then cksum < "$t/$entry" > "$t/$entry.cksum"; chmod 600 "$t/$entry.cksum"; fi
  done
done
printf 'hstack-snapshot-v1\n' > "$t/.complete"
chmod 600 "$t/.complete"
mv "$t" "$d"
[ -f "$d/.complete" ] && [ ! -d "$t" ] || { echo "snapshot: recovery-point publication failed; nothing was written." >&2; exit 1; }
t=""
REMOTE
}
# rollback: restore the exact pre-skill .env + config.yaml (so a FORCE=1 re-run that
# fails puts the previous working value back), reload the gateway with gw_reload, then retire the snapshot.
rollback() {
  local rc=0
  remote "SKILL_NAME='$SKILL_NAME' bash -s" <<'REMOTE' || rc=$?
set -e
trap 'echo "rollback: restoration failed; active snapshot retained for recovery." >&2' ERR
d="$HOME/.hermes/.hstack-backup/$SKILL_NAME"
[ ! -d "$d.creating" ] || { echo "rollback: snapshot creation is still in progress; retry later." >&2; exit 1; }
[ -d "$d" ] || { echo "rollback: no snapshot for $SKILL_NAME; nothing was written."; exit 10; }
# Incomplete legacy snapshots cannot prove that their final copy succeeded. Keep them for manual recovery.
[ -f "$d/.complete" ] && [ "$(cat "$d/.complete")" = hstack-snapshot-v1 ] \
  || { echo "rollback: snapshot is incomplete or legacy; active snapshot retained for manual recovery." >&2; exit 1; }
for f in .env config.yaml; do
  count=0
  for entry in "$f" "$f.absent"; do
    if [ -f "$d/$entry" ]; then
      count=$((count + 1))
      [ -f "$d/$entry.cksum" ] || { echo "rollback: missing checksum for $f; active snapshot retained for recovery." >&2; exit 1; }
      checksum=$(cksum < "$d/$entry")
      expected_checksum=$(cat "$d/$entry.cksum")
      [ "$checksum" = "$expected_checksum" ] \
        || { echo "rollback: corrupt backup for $f; active snapshot retained for recovery." >&2; exit 1; }
    fi
  done
  [ "$count" = 1 ] || { echo "rollback: missing or ambiguous backup for $f; active snapshot retained for recovery." >&2; exit 1; }
done
for f in .env config.yaml; do
  if [ -f "$d/$f" ]; then
    cp -p "$d/$f" "$HOME/.hermes/$f"
    cmp -s "$d/$f" "$HOME/.hermes/$f"
  elif [ -f "$d/$f.absent" ]; then
    rm -f "$HOME/.hermes/$f"
    [ ! -e "$HOME/.hermes/$f" ] && [ ! -L "$HOME/.hermes/$f" ] || { echo "rollback: file removal failed; active snapshot retained." >&2; exit 1; }
  else
    echo "rollback: missing backup for $f; active snapshot retained for recovery." >&2
    exit 1
  fi
done
[ -f "$HOME/.hermes/.env" ] && chmod 600 "$HOME/.hermes/.env"
exit 0
REMOTE
  [ "$rc" = 10 ] && return 0
  [ "$rc" = 0 ] || return "$rc"
  gw_reload >/dev/null \
    || { echo "rollback: files restored, but the gateway did NOT come back ready — active snapshot retained; inspect the gateway and retry rollback after resolving the fault" >&2; return 1; }
  finish || return 1
  echo "rollback: restored ~/.hermes/.env and config.yaml to their pre-$SKILL_NAME state."
}

# finish: mark the run successful; keep the snapshot as .prev for a manual undo.
finish() {
  remote "SKILL_NAME='$SKILL_NAME' bash -s" <<'REMOTE'
set -e
trap 'echo "finish: could not retire the active snapshot; inspect the recovery point before continuing." >&2' ERR
d="$HOME/.hermes/.hstack-backup/$SKILL_NAME"
[ -d "$d" ] || exit 0
rm -rf "$d.prev"
[ ! -e "$d.prev" ] && [ ! -L "$d.prev" ] || { echo "snapshot retirement: previous destination remains; active snapshot retained." >&2; exit 1; }
mv "$d" "$d.prev"
[ ! -e "$d" ] && [ ! -L "$d" ] && [ -f "$d.prev/.complete" ] || { echo "snapshot retirement: move could not be verified; inspect the recovery point." >&2; exit 1; }
REMOTE
}
# Reject line breaks, quote/backslash characters and controls: values also become curl config lines.
secret_line_ok() {
  [ -n "$1" ] && printf '%s' "$1" | LC_ALL=C grep -Eq '^[!-~]+$' || return 1
  case "$1" in *'"'*|*"'"*|*'\'*) return 1 ;; esac
}
# No response body or credential is logged. JSON errors on HTTP 200 are failures too.
api_check() {
  local reply
  reply=$(remote 'curl -q -fsS --connect-timeout 10 --max-time 25 --config -' 2>/dev/null) || return 1
  printf '%s' "$reply" | python3 -c 'import json,sys; d=json.load(sys.stdin); assert not (isinstance(d,dict) and (d.get("errors") or d.get("error") or d.get("success") is False))' 2>/dev/null
}
# Extra native-skill artifacts are owned separately; canonical rollback still restores env/config bytes.
restore_native_skill() {
  remote "SKILL_NAME='$SKILL_NAME' bash -s" <<'REMOTE'
set -e
d="$HOME/.hermes/.hstack-backup/$SKILL_NAME"
p="$HOME/.hermes/skills/$SKILL_NAME-runtime"
[ -f "$d/native-created" ] || exit 0


---

*Content truncated.*

When not to use it

  • →When connecting Jira Data Center/Server (not Atlassian Cloud)
  • →When the user does not have SSH access to their VPS
  • →When the user needs to connect a system other than Jira

Limitations

  • →The skill connects Jira (Atlassian Cloud).
  • →The skill connects to a self-hosted Hermes Agent over SSH.
  • →The skill uses Atlassian's official remote MCP server (Rovo).

How it compares

This skill automates the secure and idempotent integration of Jira Cloud with a self-hosted Hermes agent via SSH, handling token storage and MCP registration, unlike a manual setup that could be error-prone.

Compared to similar skills

integration-jira side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
integration-jira (this skill)04moCautionAdvanced
tapd04moReviewAdvanced
fc-assistant05moNo flagsAdvanced
linear06moNo flagsIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry