Files
Triple-C/container/entrypoint.sh
T
shadow-testandClaude Opus 5 016de8f641
Build App (Preview) / compute-version (pull_request) Successful in 3s
Build Container / build-container (pull_request) Successful in 10m5s
Build App (Preview) / create-release (pull_request) Successful in 1s
Build App (Preview) / build-macos (pull_request) Successful in 4m31s
Build App (Preview) / build-linux (pull_request) Successful in 5m21s
Build App (Preview) / build-windows (pull_request) Successful in 19m1s
Build App (Preview) / prune-previews (pull_request) Successful in 1s
Close the blockers from the fifth audit
Docs and disclosure. HOW-TO-USE.md's settings table still described the
pre-fix behaviour — and help_commands.rs fetches that file from GitHub
main at runtime, ahead of the embedded copy, so it would have reached
every user's Help dialog the moment this merged. The Config tab named
three settings that need a base-image update; there are four, and the
omitted one (Session recap) is the one that fails *without* the "won't
switch off" symptom the warning teaches. Both now also state the cost
nobody had written down: changing any of these recreates the container,
which commits a layer.

Two stale comments that told a reviewer the code was safe when it was
not. compute_claude_code_settings_fingerprint still claimed the
historical fingerprint is preserved so an upgrade cannot churn every
container — carried over from before the widening, false since the
format string changed. And capabilities/default.json, which is the
reviewed threat model of record, described a "Save to host…" action this
branch deletes.

Security and correctness. update_settings validated env vars and nothing
else, so the *global* default_ssh_key_path — the fallback for every
project without an override — took `/` and read-only bind-mounted the
host, which entrypoint.sh then copies into the home volume. classify_
mount_source ran canonicalize on the raw string, which resolves a
relative path against Triple-C's own cwd, so `.` and `..` were accepted
or refused depending on where the app was launched; the daemon then
refuses the mount and the project can never start. Its test passed only
because its examples did not exist under app/src-tauri.

bind_mount_exclusions still derived a path from every row while
project_path_mounts had learned to skip unmountable ones, so a legacy
row made /workspace/<name> ordinary container content that a migration
would then exclude from staging and destroy. The skip is also logged now
rather than silently dropping a folder.

The terminal's file-in path checked is_dir() but not file type, so a
dropped FIFO blocked forever with no timeout — and it is the only route
in now. The web terminal labelled sessions from a global set at request
time, so two quick opens swapped them; harmless until Shift+Enter became
type-dependent, at which point a mislabelled Claude session submitted a
half-written prompt. Opened now carries the type.

Every ~/.claude.json write goes through one atomic helper. The
awsAuthRefresh branches still truncated in place — the same corruption
the Shift+Enter block was fixed for twenty lines later, and its own
comment said so. Demonstrated: a failed write now leaves the original
byte-identical.

And the registration test I added yesterday could pass while the
property was false: an audit got five real unregistered commands past its
exact-string attribute match, and "exactly once" was in its name but not
its body. Mutation-checked against all six shapes.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GBq2rGum6GX7xXgsas1fDc
2026-08-23 18:45:11 -07:00

650 lines
32 KiB
Bash

#!/bin/bash
# NOTE: set -e is intentionally omitted. A failing usermod/groupmod must not
# kill the entire entrypoint — SSH setup, git config, and the final exec
# must still run so the container is usable even if remapping fails.
#
# NOTE: /home/claude is the mount point of the named volume
# triple-c-home-{projectId}, so the *image's* copy of that directory is
# seed-only: after a project's first start it is masked permanently. Anything
# this script writes under /home/claude on **every** start does reach existing
# projects (that is why the CLAUDE.md, git config and Mission Control skill
# copies are written here rather than baked into the image). Anything added to
# /home/claude in the Dockerfile reaches new projects only, forever. Put
# upgradable content in /usr/local/bin or /opt, or seed it from here.
# See "Container Lifecycle" in the repo's CLAUDE.md.
# ── UID/GID remapping ──────────────────────────────────────────────────────
# Match the container's claude user to the host user's UID/GID so that
# bind-mounted files (project dir, docker socket) have correct ownership.
remap_uid_gid() {
local target_uid="${HOST_UID}"
local target_gid="${HOST_GID}"
local current_uid
local current_gid
current_uid=$(id -u claude 2>/dev/null) || { echo "entrypoint: claude user not found"; return 1; }
current_gid=$(id -g claude 2>/dev/null) || { echo "entrypoint: claude group not found"; return 1; }
# ── GID remapping ──
if [ -n "$target_gid" ] && [ "$target_gid" != "$current_gid" ]; then
# If another group already holds the target GID, move it out of the way
local blocking_group
blocking_group=$(getent group "$target_gid" 2>/dev/null | cut -d: -f1)
if [ -n "$blocking_group" ] && [ "$blocking_group" != "claude" ]; then
echo "entrypoint: moving group '$blocking_group' from GID $target_gid to 65533"
groupmod -g 65533 "$blocking_group" || echo "entrypoint: warning — failed to relocate group '$blocking_group'"
fi
groupmod -g "$target_gid" claude \
&& echo "entrypoint: claude GID -> $target_gid" \
|| echo "entrypoint: warning — groupmod -g $target_gid claude failed"
fi
# ── UID remapping ──
if [ -n "$target_uid" ] && [ "$target_uid" != "$current_uid" ]; then
# If another user already holds the target UID, move it out of the way
local blocking_user
blocking_user=$(getent passwd "$target_uid" 2>/dev/null | cut -d: -f1)
if [ -n "$blocking_user" ] && [ "$blocking_user" != "claude" ]; then
echo "entrypoint: moving user '$blocking_user' from UID $target_uid to 65533"
usermod -u 65533 "$blocking_user" || echo "entrypoint: warning — failed to relocate user '$blocking_user'"
fi
usermod -u "$target_uid" claude \
&& echo "entrypoint: claude UID -> $target_uid" \
|| echo "entrypoint: warning — usermod -u $target_uid claude failed"
fi
}
remap_uid_gid
# Fix ownership of home directory after UID/GID change
chown -R claude:claude /home/claude
# ── Corporate CA certificates ───────────────────────────────────────────────
# The host's CA material is bind-mounted read-only at /tmp/.host-ca. Triple-C
# mounts a *directory* as-is and a *single file* as /tmp/.host-ca/<name>.crt,
# so this only ever has to deal with a directory (the file branch below is
# defensive).
#
# Runs before everything that touches the network — the git credential helper,
# ssh-keyscan, and especially the `claude update` at the bottom of this file,
# which is itself an HTTPS call that fails behind a TLS-terminating proxy
# without this.
#
# Two things are easy to get wrong here:
# 1. `update-ca-certificates` globs /usr/local/share/ca-certificates/*.crt
# case-sensitively. A `.pem` that is merely copied in is ignored in total
# silence, so certificates are *renamed*, not copied.
# 2. Chrome/Chromium read neither /etc/ssl nor $SSL_CERT_FILE; they have
# their own NSS database at ~/.pki/nssdb, seeded below with certutil.
#
# NODE_EXTRA_CA_CERTS / REQUESTS_CA_BUNDLE / SSL_CERT_FILE are deliberately NOT
# exported here. Every terminal is a separate `docker exec`, which inherits the
# container's configured env and sees nothing this script exported — the same
# reason $BROWSER had to become an image-level ENV. Triple-C sets them on the
# container at creation time instead. (They are forwarded into the cron
# environment file further down, because cron jobs start from a bare env.)
CA_SRC="/tmp/.host-ca"
CA_STORE="/usr/local/share/ca-certificates"
CA_PREFIX="triple-c-"
CA_BUNDLE="/etc/ssl/certs/ca-certificates.crt"
CA_STAMP="/var/lib/triple-c/ca.stamp"
CA_NSSDB="/home/claude/.pki/nssdb"
# Mirror of `container_cert_name()` in app/src-tauri/src/docker/ca_certs.rs.
# The two must agree; the Rust side has the unit tests.
ca_normalise_name() {
local name stem
name=$(printf '%s' "$1" | tr -c 'A-Za-z0-9._-' '_')
while [ "${name#.}" != "$name" ]; do name="${name#.}"; done
stem="${name%.*}"
[ -z "$stem" ] && stem="corporate-ca"
printf '%s.crt' "$stem"
}
ca_source_files() {
if [ -d "$CA_SRC" ]; then
find "$CA_SRC" -maxdepth 1 -type f \
\( -iname '*.crt' -o -iname '*.pem' -o -iname '*.cer' \
-o -iname '*.cert' -o -iname '*.ca-bundle' \) 2>/dev/null | sort
elif [ -f "$CA_SRC" ]; then
printf '%s\n' "$CA_SRC"
fi
}
# Seed Chrome/Chromium's NSS database. Tolerant by design: a missing certutil
# or a broken profile must warn, never fail the container start.
# ~/.pki lives in the home volume, so this persists once done; the system store
# lives in the writable layer and is re-applied on every start.
ca_seed_nssdb() {
if ! command -v certutil >/dev/null 2>&1; then
echo "entrypoint: warning — certutil not found (install libnss3-tools); Chrome/Chromium in this container will not trust the corporate CA"
return 0
fi
su -s /bin/bash claude -c '
db="$HOME/.pki/nssdb"
mkdir -p "$db" || exit 1
if [ ! -f "$db/cert9.db" ]; then
certutil -d "sql:$db" -N --empty-password >/dev/null 2>&1 || exit 1
fi
for f in /usr/local/share/ca-certificates/triple-c-*.crt; do
[ -f "$f" ] || continue
nick="triple-c:$(basename "$f" .crt)"
# Delete first so re-running replaces rather than duplicates.
certutil -d "sql:$db" -D -n "$nick" >/dev/null 2>&1
certutil -d "sql:$db" -A -t "C,," -n "$nick" -i "$f" >/dev/null 2>&1 \
|| echo "entrypoint: warning — certutil could not add $nick"
done
' && echo "entrypoint: seeded Chrome/Chromium NSS database with the corporate CA" \
|| echo "entrypoint: warning — NSS database seeding failed (continuing)"
}
install_corporate_ca() {
local files fp stamp f base name count installed
files=$(ca_source_files)
if [ -z "$files" ]; then
# Nothing configured — but /usr/local/share is in the writable layer and
# `docker commit` bakes it into the project's snapshot image, so a cert
# installed by a previous configuration would ride that snapshot into
# every future container. Turning the setting off has to actively undo.
if ls "$CA_STORE/$CA_PREFIX"*.crt >/dev/null 2>&1; then
echo "entrypoint: removing previously installed corporate CA certificates"
rm -f "$CA_STORE/$CA_PREFIX"*.crt
update-ca-certificates --fresh >/dev/null 2>&1 \
|| echo "entrypoint: warning — update-ca-certificates failed while removing certificates"
rm -f "$CA_STAMP"
fi
if [ -e "$CA_SRC" ]; then
echo "entrypoint: warning — $CA_SRC holds no certificate files"
fi
return 0
fi
# Idempotent and cheap: the certs are already installed on a plain restart
# (the writable layer survives stop/start), so hash the sources and skip the
# work when nothing has moved. The NSS database is checked separately
# because it lives in the home volume and can be wiped independently.
fp=$(printf '%s\n' "$files" | xargs -d '\n' -r sha256sum 2>/dev/null | sha256sum | cut -d' ' -f1)
stamp=$(cat "$CA_STAMP" 2>/dev/null)
if [ -n "$fp" ] && [ "$fp" = "$stamp" ] && [ -s "$CA_BUNDLE" ]; then
if [ -f "$CA_NSSDB/cert9.db" ]; then
echo "entrypoint: corporate CA certificates already installed"
return 0
fi
ca_seed_nssdb
return 0
fi
mkdir -p "$CA_STORE" "$(dirname "$CA_STAMP")"
rm -f "$CA_STORE/$CA_PREFIX"*.crt
installed=0
while IFS= read -r f; do
[ -n "$f" ] || continue
base=$(basename "$f")
name="$CA_PREFIX$(ca_normalise_name "$base")"
count=$(grep -c -- '-----BEGIN CERTIFICATE-----' "$f" 2>/dev/null || true)
[ -n "$count" ] || count=0
if [ "$count" -gt 1 ]; then
# A corporate trust chain is usually delivered as one PEM holding
# root + intermediates. update-ca-certificates handles exactly one
# certificate per file, so split it.
awk -v out="$CA_STORE/${name%.crt}" '
/-----BEGIN CERTIFICATE-----/ { n++; f = out "-" n ".crt" }
n > 0 { print > f }
' "$f" && installed=$((installed + count))
elif [ "$count" -eq 1 ]; then
cp -f "$f" "$CA_STORE/$name" && installed=$((installed + 1))
else
echo "entrypoint: warning — $f holds no PEM certificate (DER is not supported), skipping"
fi
done <<< "$files"
chmod 644 "$CA_STORE/$CA_PREFIX"*.crt 2>/dev/null
if [ "$installed" -eq 0 ]; then
echo "entrypoint: warning — no usable certificates found under $CA_SRC"
return 0
fi
if update-ca-certificates >/dev/null 2>&1; then
echo "entrypoint: installed $installed corporate CA certificate(s) into the system trust store"
printf '%s' "$fp" > "$CA_STAMP"
else
echo "entrypoint: warning — update-ca-certificates failed; corporate certificates may not be trusted"
fi
ca_seed_nssdb
}
install_corporate_ca
# ── SSH key setup ──────────────────────────────────────────────────────────
# Host SSH dir is mounted read-only at /tmp/.host-ssh.
# Copy to /home/claude/.ssh so we can fix permissions.
if [ -d /tmp/.host-ssh ]; then
rm -rf /home/claude/.ssh
cp -a /tmp/.host-ssh /home/claude/.ssh
chown -R claude:claude /home/claude/.ssh
chmod 700 /home/claude/.ssh
find /home/claude/.ssh -type f -name "id_*" ! -name "*.pub" -exec chmod 600 {} \;
find /home/claude/.ssh -type f -name "*.pub" -exec chmod 644 {} \;
if [ -f /home/claude/.ssh/known_hosts ]; then
chmod 644 /home/claude/.ssh/known_hosts
fi
if [ -f /home/claude/.ssh/config ]; then
chmod 600 /home/claude/.ssh/config
fi
fi
# Append common host keys (avoid duplicates)
su -s /bin/bash claude -c '
mkdir -p /home/claude/.ssh
ssh-keyscan -t ed25519,rsa github.com gitlab.com bitbucket.org >> /home/claude/.ssh/known_hosts 2>/dev/null || true
sort -u -o /home/claude/.ssh/known_hosts /home/claude/.ssh/known_hosts
'
# ── AWS config setup ──────────────────────────────────────────────────────────
# Host AWS dir is mounted read-only at /tmp/.host-aws.
# Copy to /home/claude/.aws so AWS CLI can write to sso/cache and cli/cache.
if [ -d /tmp/.host-aws ]; then
rm -rf /home/claude/.aws
cp -a /tmp/.host-aws /home/claude/.aws
chown -R claude:claude /home/claude/.aws
chmod 700 /home/claude/.aws
# Ensure writable cache directories exist
mkdir -p /home/claude/.aws/sso/cache /home/claude/.aws/cli/cache
chown -R claude:claude /home/claude/.aws/sso /home/claude/.aws/cli
# Inline sso_session properties into profile sections so AWS SDKs that don't
# support the sso_session indirection format can resolve sso_region, etc.
if [ -f /home/claude/.aws/config ]; then
python3 -c '
import configparser, sys
c = configparser.ConfigParser()
c.read(sys.argv[1])
for sec in c.sections():
if not sec.startswith("profile ") and sec != "default":
continue
session = c.get(sec, "sso_session", fallback=None)
if not session or c.has_option(sec, "sso_start_url"):
continue
ss = f"sso-session {session}"
if not c.has_section(ss):
continue
for key in ("sso_start_url", "sso_region", "sso_registration_scopes"):
val = c.get(ss, key, fallback=None)
if val:
c.set(sec, key, val)
with open(sys.argv[1], "w") as f:
c.write(f)
' /home/claude/.aws/config 2>/dev/null || true
fi
fi
# ── Git credential helper (for HTTPS token) ─────────────────────────────────
if [ -n "$GIT_TOKEN" ]; then
CRED_FILE="/home/claude/.git-credentials"
: > "$CRED_FILE"
chmod 600 "$CRED_FILE"
chown claude:claude "$CRED_FILE"
echo "https://oauth2:${GIT_TOKEN}@github.com" >> "$CRED_FILE"
echo "https://oauth2:${GIT_TOKEN}@gitlab.com" >> "$CRED_FILE"
echo "https://oauth2:${GIT_TOKEN}@bitbucket.org" >> "$CRED_FILE"
git config --file /home/claude/.gitconfig credential.helper "store --file=$CRED_FILE"
unset GIT_TOKEN
fi
# ── Git user config ──────────────────────────────────────────────────────────
if [ -n "$GIT_USER_NAME" ]; then
git config --file /home/claude/.gitconfig user.name "$GIT_USER_NAME"
fi
if [ -n "$GIT_USER_EMAIL" ]; then
git config --file /home/claude/.gitconfig user.email "$GIT_USER_EMAIL"
fi
chown claude:claude /home/claude/.gitconfig 2>/dev/null || true
# ── Claude instructions ──────────────────────────────────────────────────────
if [ -n "$CLAUDE_INSTRUCTIONS" ]; then
mkdir -p /home/claude/.claude
printf '%s\n' "$CLAUDE_INSTRUCTIONS" > /home/claude/.claude/CLAUDE.md
chown claude:claude /home/claude/.claude/CLAUDE.md
unset CLAUDE_INSTRUCTIONS
fi
# ── Mission Control setup ───────────────────────────────────────────────────
if [ "$MISSION_CONTROL_ENABLED" = "1" ]; then
MC_HOME="/home/claude/mission-control"
MC_LINK="/workspace/mission-control"
if [ ! -d "$MC_HOME" ]; then
echo "entrypoint: installing mission-control..."
cp -r /opt/mission-control "$MC_HOME"
chown -R claude:claude "$MC_HOME"
else
echo "entrypoint: mission-control already present, skipping install"
fi
# Symlink into workspace so Claude sees it at /workspace/mission-control
ln -sfn "$MC_HOME" "$MC_LINK"
chown -h claude:claude "$MC_LINK"
# Install skills to ~/.claude/skills/ so Claude Code discovers them automatically
if [ -d "$MC_HOME/.claude/skills" ]; then
mkdir -p /home/claude/.claude/skills
cp -r "$MC_HOME/.claude/skills/"* /home/claude/.claude/skills/ 2>/dev/null
chown -R claude:claude /home/claude/.claude/skills
echo "entrypoint: mission-control skills installed to ~/.claude/skills/"
fi
unset MISSION_CONTROL_ENABLED
fi
# ── Feature skills ──────────────────────────────────────────────────────────
# Skills owned by a Triple-C feature rather than by Mission Control. Installed
# when the feature is on, removed when it is off: ~/.claude is a persisted
# volume, so a skill left behind after its feature is disabled would keep
# telling an agent to use a capability the container no longer has.
#
# Copied on every start rather than only when absent, so a fix to a skill
# reaches projects that already have the old copy. Local edits under these
# directories do not survive — treat /opt/triple-c-skills as the source.
#
# The source lives in the *base image*, so a project whose container predates it
# recreates from its own snapshot and has no /opt/triple-c-skills to copy from.
# That case says so rather than returning silently: the toggle is on, the
# capability is there, and the skill simply never appears — which is impossible
# to work out from the outside.
install_feature_skill() {
local _name="$1"
local _enabled="$2"
local _src="/opt/triple-c-skills/$1"
local _dest="/home/claude/.claude/skills/$1"
# Reject anything that is not a plain directory name. The disabled branch
# `rm -rf`s $_dest under a *persisted volume*, so a blank name would take the
# whole skills directory (Mission Control's included) and `../x` would escape
# it entirely. Only the literal `pia-vpn` is passed today; this is so that
# stays true.
case "$_name" in
''|*/*|.*) echo "entrypoint: install_feature_skill: bad skill name '$_name'"; return 1 ;;
esac
if [ "$_enabled" = "1" ]; then
if [ ! -d "$_src" ]; then
echo "entrypoint: $_name skill unavailable — this container's base image predates it; migrate the project to get it"
return 0
fi
# Checked, not assumed: with no `set -e` in this script every step here
# can fail (full volume, read-only mount, a file where the directory
# should be) and the success line would still print.
mkdir -p /home/claude/.claude/skills || {
echo "entrypoint: $_name skill install FAILED (cannot create ~/.claude/skills)"; return 1; }
# Not just $_dest: when Mission Control is off nothing else creates the
# parent, so root would own it and `claude` could not add a skill there.
chown claude:claude /home/claude/.claude/skills
# Stage then swap. Copying over the live path meant a failure (full
# volume, read-only mount) left a truncated SKILL.md and no script
# behind, root-owned, on a persisted volume — which Claude Code then
# discovers and loads.
rm -rf "$_dest.new"
cp -r "$_src" "$_dest.new" || {
rm -rf "$_dest.new"
echo "entrypoint: $_name skill install FAILED (copy from $_src); previous copy left intact"
return 1; }
chown -R claude:claude "$_dest.new"
rm -rf "$_dest"
mv "$_dest.new" "$_dest"
echo "entrypoint: $_name skill installed to ~/.claude/skills/"
elif [ -e "$_dest" ] || [ -L "$_dest" ]; then
# -e/-L rather than -d: a leftover *file* at that path must go too.
rm -rf "$_dest"
echo "entrypoint: $_name skill removed (feature disabled)"
fi
}
install_feature_skill pia-vpn "${VPN_SUPPORT_ENABLED:-0}"
unset VPN_SUPPORT_ENABLED
# ── Claude Code settings ────────────────────────────────────────────────────
# Apply the managed Claude Code settings to ~/.claude/settings.json, keeping
# every key the user set inside the container.
#
# `settings.json` lives on the persisted triple-c-claude-config-{id} volume, so
# it outlives the container and a plain `.[0] * .[1]` merge could only ever
# *add*. That is what made every one of these settings one-way: switching one
# off in Triple-C omitted its key, the merge preserved the old on-value, and the
# setting stayed on until a destructive Reset. So Rust states the whole managed
# key set on every start, in two parts: CLAUDE_CODE_SETTINGS_JSON holds the keys
# that have a value, and CLAUDE_CODE_SETTINGS_CLEAR is a JSON array of the key
# names whose neutral state is *unset* (`tui`, `effortLevel`, `viewMode`,
# `awaySummaryEnabled`) and which must therefore be deleted rather than pinned
# to a stand-in value.
#
# The two are kept apart rather than using a null-means-delete payload because
# a project recreates from its own snapshot image, which carries whatever
# entrypoint it was built with. An older one merges with a plain `.[0] * .[1]`
# and would write literal nulls into the user's settings.json; it ignores
# CLAUDE_CODE_SETTINGS_CLEAR instead, keeping the old sticky behaviour until the
# project is migrated or Reset.
# See `build_claude_code_settings_json` in app/src-tauri/src/docker/container.rs.
if [ -n "$CLAUDE_CODE_SETTINGS_JSON" ]; then
SETTINGS_FILE="/home/claude/.claude/settings.json"
mkdir -p /home/claude/.claude
# One code path for "file exists" and "file doesn't".
[ -f "$SETTINGS_FILE" ] || printf '{}\n' > "$SETTINGS_FILE"
MERGED=$(jq -s --argjson clear "${CLAUDE_CODE_SETTINGS_CLEAR:-[]}" '
.[0] as $current
| .[1] as $set
| ($current * $set) | delpaths($clear | map([.]))
' "$SETTINGS_FILE" <(printf '%s' "$CLAUDE_CODE_SETTINGS_JSON") 2>/dev/null)
if [ -n "$MERGED" ]; then
printf '%s\n' "$MERGED" > "$SETTINGS_FILE"
else
echo "entrypoint: warning — failed to merge Claude Code settings into $SETTINGS_FILE"
fi
chown claude:claude "$SETTINGS_FILE"
chmod 600 "$SETTINGS_FILE"
unset CLAUDE_CODE_SETTINGS_JSON
unset CLAUDE_CODE_SETTINGS_CLEAR
fi
# ── AWS SSO auth refresh command ──────────────────────────────────────────────
# When set (Bedrock + profile/SSO auth), inject awsAuthRefresh into
# ~/.claude.json so Claude Code calls triple-c-sso-refresh when AWS credentials
# expire mid-session. When NOT set, strip any awsAuthRefresh left behind by a
# previous Bedrock-profile session — ~/.claude.json lives in the persisted home
# volume, so without this the container keeps trying to run the SSO refresh even
# after switching to a non-SSO backend (Anthropic/Ollama) or to static creds.
# Replace ~/.claude.json atomically: write a sibling temp file, then rename.
#
# `> "$CLAUDE_JSON"` truncates before it writes, so a write that fails part-way
# — a full home volume being the obvious way, and bounding that volume is what
# half this release is about — leaves the file unparseable. It holds the OAuth
# account, and the damage does not self-heal: the next start's `jq` fails on the
# corrupt file, `MERGED` comes back empty, and the `[ -n "$MERGED" ]` guard
# skips the write that would have repaired it. `triple-c-task-runner` has done
# it this way all along.
write_claude_json() {
_wcj_tmp="${CLAUDE_JSON}.triple-c-tmp"
if printf '%s\n' "$1" > "$_wcj_tmp" 2>/dev/null; then
mv -f "$_wcj_tmp" "$CLAUDE_JSON" 2>/dev/null || rm -f "$_wcj_tmp"
else
rm -f "$_wcj_tmp"
echo "entrypoint: warning — could not write $CLAUDE_JSON (leaving it as it was)"
return 1
fi
# By name, after the rename, so these land on the new inode.
chown claude:claude "$CLAUDE_JSON"
chmod 600 "$CLAUDE_JSON"
}
CLAUDE_JSON="/home/claude/.claude.json"
if [ -n "$AWS_SSO_AUTH_REFRESH_CMD" ]; then
if [ -f "$CLAUDE_JSON" ]; then
MERGED=$(jq --arg cmd "$AWS_SSO_AUTH_REFRESH_CMD" '.awsAuthRefresh = $cmd' "$CLAUDE_JSON" 2>/dev/null)
if [ -n "$MERGED" ]; then
write_claude_json "$MERGED"
fi
else
# No existing file, so there is nothing to destroy — but go through the
# same helper so the owner and mode are set in one place.
write_claude_json "$(printf '{"awsAuthRefresh":"%s"}' "$AWS_SSO_AUTH_REFRESH_CMD")"
fi
unset AWS_SSO_AUTH_REFRESH_CMD
elif [ -f "$CLAUDE_JSON" ] && grep -q '"awsAuthRefresh"' "$CLAUDE_JSON" 2>/dev/null; then
# Only rewrite when the key is actually present, to avoid a needless jq
# reformat of ~/.claude.json on every start of a non-SSO backend.
MERGED=$(jq 'del(.awsAuthRefresh)' "$CLAUDE_JSON" 2>/dev/null)
if [ -n "$MERGED" ]; then
write_claude_json "$MERGED"
fi
fi
# ── Shift+Enter key binding ───────────────────────────────────────────────────
# Triple-C's terminals send ESC+CR (`\e\r`) on Shift+Enter — the same bytes
# `/terminal-setup` installs for VS Code, Cursor, Alacritty and Zed — and Claude
# Code decodes those unconditionally. What it cannot see is that the binding is
# already in place, so it keeps printing its "run /terminal-setup" tip. This
# flag is what that tip is gated on: purely cosmetic, and it changes nothing
# about how the sequence is decoded.
#
# $CLAUDE_JSON is already set by the awsAuthRefresh block above.
if [ -f "$CLAUDE_JSON" ]; then
# Only rewrite when the value isn't already true, to avoid a needless jq
# reformat of ~/.claude.json on every single start.
if ! grep -q '"shiftEnterKeyBindingInstalled"[[:space:]]*:[[:space:]]*true' "$CLAUDE_JSON" 2>/dev/null; then
# Atomic, via `write_claude_json` — see its comment for why a plain
# `>` on this file can permanently destroy the OAuth login.
MERGED=$(jq '.shiftEnterKeyBindingInstalled = true' "$CLAUDE_JSON" 2>/dev/null)
if [ -n "$MERGED" ]; then
write_claude_json "$MERGED"
fi
fi
else
# Nothing to destroy, but the helper owns the owner/mode too.
write_claude_json '{"shiftEnterKeyBindingInstalled":true}'
fi
# ── Docker socket permissions ────────────────────────────────────────────────
if [ -S /var/run/docker.sock ]; then
DOCKER_GID=$(stat -c '%g' /var/run/docker.sock)
if ! getent group "$DOCKER_GID" > /dev/null 2>&1; then
groupadd -g "$DOCKER_GID" docker-host
fi
DOCKER_GROUP=$(getent group "$DOCKER_GID" | cut -d: -f1)
usermod -aG "$DOCKER_GROUP" claude
fi
# ── Timezone setup ───────────────────────────────────────────────────────────
if [ -n "${TZ:-}" ]; then
if [ -f "/usr/share/zoneinfo/$TZ" ]; then
ln -sf "/usr/share/zoneinfo/$TZ" /etc/localtime
echo "$TZ" > /etc/timezone
echo "entrypoint: timezone set to $TZ"
else
echo "entrypoint: warning — timezone '$TZ' not found in /usr/share/zoneinfo"
fi
fi
# ── Browser / URL relay ──────────────────────────────────────────────────────
# Tools that open a browser (gh, aws sso login, gcloud, python webbrowser, ...)
# consult $BROWSER first. Point it at the relay shim, which forwards the URL to
# the host's browser over the terminal. The image already sets this as an ENV —
# that is what `docker exec` terminal sessions inherit — but exporting it here
# means the entrypoint's own children see it too, and (below) that it is
# captured into the cron environment file for scheduled tasks. Under cron there
# is no terminal, so the shim degrades to printing the URL into the task log.
if [ -x /usr/local/bin/triple-c-open ]; then
export BROWSER=/usr/local/bin/triple-c-open
fi
# ── Playwright browser config ───────────────────────────────────────────────
# Seed ~/.playwright/cli.config.json on every start.
#
# Without it `playwright-cli` resolves to channel `chrome` — system Google
# Chrome — with the Chromium sandbox ON, and these containers do not permit
# unprivileged user namespaces, so the browser aborts with "Failed to move to
# new namespace ... Operation not permitted". On a base image that no longer
# ships Google Chrome the same default fails the other way, with "Chromium
# distribution 'chrome' is not found". One cause, two error messages, and
# neither of them looks like a configuration problem.
#
# Seeded here rather than baked into the image because ~/.playwright is inside
# the home volume: an image copy would reach new projects only, and every
# existing project would stay broken forever. Written on every start from a
# source outside the volume, the way CLAUDE_INSTRUCTIONS and the Mission
# Control skills already are.
#
# --seed-config-only is the cheap path: no npm install, no browser download, no
# apt, no verify launch, nothing over the network. It writes one small file if
# it is absent and returns. Measured at ~2 ms. The heavier repairs stay
# on-demand — run `triple-c-playwright-heal` with no arguments for those.
if [ -x /usr/local/bin/triple-c-playwright-heal ]; then
/usr/local/bin/triple-c-playwright-heal --seed-config-only --quiet || \
echo "entrypoint: warning — playwright config seeding failed (browser view may not launch)"
fi
# ── Scheduler setup ─────────────────────────────────────────────────────────
SCHEDULER_DIR="/home/claude/.claude/scheduler"
mkdir -p "$SCHEDULER_DIR/tasks" "$SCHEDULER_DIR/logs" "$SCHEDULER_DIR/notifications"
chown -R claude:claude "$SCHEDULER_DIR"
# Start cron daemon (runs as root, executes jobs per user crontab)
cron
# Save environment variables for cron jobs (cron runs with a minimal env)
#
# HOME is deliberately NOT captured here. This entrypoint runs as root, so the
# snapshot would record HOME=/root — and the task runner sources this file with
# `set -a`, which would overwrite the HOME cron gives the job. Claude Code then
# looks for its OAuth credential at /root/.claude/.credentials.json instead of
# /home/claude/.claude/.credentials.json and every scheduled task dies with
# "Not logged in · Please run /login". Cron still needs a HOME, so it is written
# explicitly below with the value the `claude` user actually has.
ENV_FILE="$SCHEDULER_DIR/.env"
: > "$ENV_FILE"
env | while IFS='=' read -r key value; do
case "$key" in
ANTHROPIC_*|AWS_*|CLAUDE_CODE_*|TRIPLE_C_PERMISSION_MODE|PATH|LANG|TZ|COLORTERM|BROWSER|NODE_EXTRA_CA_CERTS|REQUESTS_CA_BUNDLE|SSL_CERT_FILE)
# Escape single quotes in value and write as KEY='VALUE'
escaped_value=$(printf '%s' "$value" | sed "s/'/'\\\\''/g")
printf "%s='%s'\n" "$key" "$escaped_value" >> "$ENV_FILE"
;;
esac
done
printf "HOME='/home/claude'\n" >> "$ENV_FILE"
chown claude:claude "$ENV_FILE"
chmod 600 "$ENV_FILE"
# Restore crontab from persisted task JSON files (survives container recreation)
if ls "$SCHEDULER_DIR/tasks/"*.json >/dev/null 2>&1; then
CRON_TMP=$(mktemp)
echo "# Triple-C scheduled tasks — managed by triple-c-scheduler" > "$CRON_TMP"
echo "# Do not edit manually; changes will be overwritten." >> "$CRON_TMP"
echo "" >> "$CRON_TMP"
for task_file in "$SCHEDULER_DIR/tasks/"*.json; do
[ -f "$task_file" ] || continue
enabled=$(jq -r '.enabled' "$task_file")
[ "$enabled" = "true" ] || continue
schedule=$(jq -r '.schedule' "$task_file")
id=$(jq -r '.id' "$task_file")
echo "$schedule /usr/local/bin/triple-c-task-runner $id" >> "$CRON_TMP"
done
crontab -u claude "$CRON_TMP" 2>/dev/null || true
rm -f "$CRON_TMP"
echo "entrypoint: restored crontab from persisted tasks"
fi
# ── Claude Code self-update ──────────────────────────────────────────────────
# Update the Claude Code CLI to the latest version on container start, before
# any terminal session launches `claude`. Runs as the claude user (the CLI is
# installed under /home/claude/.claude/bin). Non-fatal and time-bounded so a
# slow or offline network never blocks container readiness.
echo "entrypoint: checking for Claude Code updates..."
timeout 120 su -s /bin/bash claude -c 'export PATH="/home/claude/.claude/bin:/home/claude/.local/bin:$PATH"; claude update' \
&& echo "entrypoint: Claude Code is up to date" \
|| echo "entrypoint: warning — Claude Code update skipped or failed (continuing)"
# ── Stay alive as claude ─────────────────────────────────────────────────────
echo "Triple-C container ready."
exec su -s /bin/bash claude -c "exec sleep infinity"