smart_ide/services/ia_dev/git-issues/agent-loop-n-cycles.sh
Nicolas Cantu 58cc2493e5 chore: consolidate ia_dev module, sync tooling, and harden gateways (0.0.5)
Initial state:
- ia_dev was historically referenced as ./ia_dev in docs and integrations, while the vendored module lives under services/ia_dev.
- AnythingLLM sync and hook installation had error masking / weak exit signaling.
- Proxy layers did not validate proxy path segments, allowing path normalization tricks.

Motivation:
- Make the IDE-oriented workflow usable (sync -> act -> deploy/preview) with explicit errors.
- Reduce security footguns in proxying and script automation.

Resolution:
- Standardize IA_DEV_ROOT usage and documentation to services/ia_dev.
- Add SSH remote data mirroring + optional AnythingLLM ingestion.
- Extend AnythingLLM pull sync to support upload-all/prefix and fail on upload errors.
- Harden smart-ide-sso-gateway and smart-ide-global-api proxying with safe-path checks and non-leaking error responses.
- Improve ia-dev-gateway runner validation and reduce sensitive path leakage.
- Add site scaffold tool (Vite/React) with OIDC + chat via sso-gateway -> orchestrator.

Root cause:
- Historical layout changes (submodule -> vendored tree) and missing central contracts for path resolution.
- Missing validation for proxy path traversal patterns.
- Overuse of silent fallbacks (|| true, exit 0 on partial failures) in automation scripts.

Impacted features:
- Project sync: git pull + AnythingLLM sync + remote data mirror ingestion.
- Site frontends: SSO gateway proxy and orchestrator intents (rag.query, chat.local).
- Agent execution: ia-dev-gateway script runner and SSE output.

Code modified:
- scripts/remote-data-ssh-sync.sh
- scripts/anythingllm-pull-sync/sync.mjs
- scripts/install-anythingllm-post-merge-hook.sh
- cron/git-pull-project-clones.sh
- services/smart-ide-sso-gateway/src/server.ts
- services/smart-ide-global-api/src/server.ts
- services/smart-ide-orchestrator/src/server.ts
- services/ia-dev-gateway/src/server.ts
- services/ia_dev/tools/site-generate.sh

Documentation modified:
- docs/** (architecture, API docs, ia_dev module + integration, scripts)

Configurations modified:
- config/services.local.env.example
- services/*/.env.example

Files in deploy modified:
- services/ia_dev/deploy/*

Files in logs impacted:
- logs/ia_dev.log (runtime only)
- .logs/* (runtime only)

Databases and other sources modified:
- None

Off-project modifications:
- None

Files in .smartIde modified:
- .smartIde/agents/*.md
- services/ia_dev/.smartIde/**

Files in .secrets modified:
- None

New patch version in VERSION:
- 0.0.5

CHANGELOG.md updated:
- yes
2026-04-04 18:36:43 +02:00

73 lines
3.3 KiB
Bash
Executable File

#!/usr/bin/env bash
# Run N cycles of (retrieval + treatment via Cursor Agent CLI + sleep 60s) in foreground.
# Uses same lock as section 2. When AGENT_LOOP_RUN_AGENT=1 and "agent" in PATH, runs
# git-issues-process workflow via CLI each cycle (same as agent-loop.sh). No contournement:
# /agent-loop 600 = agent runs this script with N=600 and timeout ~11h; script does 600 full cycles.
#
# Usage: from ia_dev root: ./git-issues/agent-loop-n-cycles.sh [N]
# N = number of cycles (default 600). Each cycle = retrieval + agent CLI (if configured) + 60s sleep.
#
set -euo pipefail
if [ -n "${HOME:-}" ] && [ -r "$HOME/.bashrc" ]; then
set +u
# shellcheck source=/dev/null
source "$HOME/.bashrc" 2>/dev/null || true
set -u
fi
[ -n "${HOME:-}" ] && [ -d "$HOME/.local/bin" ] && export PATH="$HOME/.local/bin:$PATH"
GIT_ISSUES_DIR="${GIT_ISSUES_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}"
ROOT="$(cd "${GIT_ISSUES_DIR}/../.." && pwd)"
IA_DEV_ROOT="${GIT_ISSUES_DIR}/.."
export GIT_ISSUES_DIR
export REPO_ROOT="${ROOT}"
cd "$ROOT"
# shellcheck source=lib.sh
source "${GIT_ISSUES_DIR}/lib.sh" 2>/dev/null || true
AGENT_LOOP_ENV="${IA_DEV_ROOT}/.secrets/git-issues/agent-loop.env"
if [ -r "$AGENT_LOOP_ENV" ]; then
set +u
# shellcheck source=/dev/null
source "$AGENT_LOOP_ENV"
set -u
fi
N="${1:-600}"
if ! [[ "$N" =~ ^[0-9]+$ ]] || [ "$N" -lt 1 ]; then
echo "[agent-loop-n-cycles] Usage: $0 [N] (N positive integer, default 600)" >&2
exit 1
fi
"${GIT_ISSUES_DIR}/agent-loop-lock-acquire.sh" || exit 1
AGENT_PROMPT="Exécute le workflow mails du spooler (agent git-issues-process). 1) Lister les mails à traiter avec ./git-issues/list-pending-spooler.sh (ou utiliser les chemins dans agent-loop.pending). 2) Pour chaque fichier .pending : lire le JSON (from, to, subject, body, message_id, base). Répondre à l'expéditeur (--to <from> du JSON), pas à une adresse fixe ; le « to » du mail reçu a déjà déterminé le projet. Rédiger une réponse pertinente (uniquement ton texte, pas de citation ; mail-send-reply.sh refuse si le body contient From:, Message-ID, wrote:, etc.). Envoyer avec ./git-issues/mail-send-reply.sh --to <from> --subject \"Re: ...\" --body \"<ta_réponse>\" --in-reply-to \"<message_id>\". Après envoi réussi : ./git-issues/write-response-spooler.sh --base <base> --to <from> --subject \"Re: ...\" --body \"<ta_réponse>\" --in-reply-to \"<message_id>\". Ne pas appeler mail-mark-read.sh (spooler)."
i=1
while [ "$i" -le "$N" ]; do
if "${GIT_ISSUES_DIR}/agent-loop-stop-requested.sh" 2>/dev/null; then
echo "[agent-loop-n-cycles] $(date -Iseconds) — Arrêt demandé (cycle $i/$N). Libération du lock."
"${GIT_ISSUES_DIR}/agent-loop-lock-release.sh"
exit 0
fi
echo "[agent-loop-n-cycles] $(date -Iseconds) — Cycle $i/$N"
"${GIT_ISSUES_DIR}/agent-loop-retrieval-once.sh" || true
if [ "${AGENT_LOOP_RUN_AGENT:-0}" = "1" ] && command -v agent >/dev/null 2>&1; then
AGENT_MODEL="${AGENT_LOOP_MODEL:-sonnet-4.6}"
if agent -p "$AGENT_PROMPT" -f --model "$AGENT_MODEL" 2>&1; then
: # agent done
else
echo "[agent-loop-n-cycles] $(date -Iseconds) — Agent terminé avec erreur ou interruption (cycle $i/$N)"
fi
fi
if [ "$i" -lt "$N" ]; then
sleep 60
fi
i=$((i + 1))
done
"${GIT_ISSUES_DIR}/agent-loop-lock-release.sh"
echo "[agent-loop-n-cycles] $(date -Iseconds)$N cycles terminés."