Command reference

For an existing WSL deployment, these scripts live in /opt/devsecops-ai/scripts. Run cd /opt/devsecops-ai before the relative commands below. Read the complete CLI walkthrough for examples, model routing, and execution behavior.

Entry point	Shell	Purpose / scope
`scripts/ask.sh`	WSL Bash	Interactive API query; reads environment file and saves thread state
`scripts/devai.sh`	WSL Bash	Development CLI, including memory/ingestion modes
`scripts/plan_exec.sh`	WSL Bash	Planning/execution workflow; choose mode deliberately
`scripts/wsl-manage.ps1`	Windows PowerShell	Control an existing WSL installation
`scripts/validate-router.sh`	WSL Bash	Registry, syntax, and routing checks; requires local dependencies
`scripts/validate-catalogs.py`	Project Python	Prompt/playbook catalog validation
`scripts/backup.sh`	WSL Bash	Partial backup; see coverage limitations
`scripts/deploy-to-wsl.sh`	WSL Bash	State-changing deployment; known revision/rollback limitations

CLI settings and paths

Setting	Script	Meaning / default
`DEVAI_API_URL`	All three	Backend URL; `http://localhost:8000`
`MODEL_SELECTION`	ask.sh	Explicit model request; empty means automatic
`TASK_TYPE`	ask.sh	Optional task hint
`DEPLOY_PATH`	ask.sh	Looks for `.env` here first; `/opt/devsecops-ai`
`DEVAI_MODEL`	devai.sh	Explicit model request; empty means automatic
`DEVAI_TASK_TYPE`	devai.sh	Optional task hint
`DEVAI_API_KEY`	devai.sh, plan_exec.sh	Optional key override, otherwise read the script’s project-root `.env`
`DEVAI_PYTHON`	devai.sh	Memory helper interpreter; project `.venv/bin/python`
`PLAN_MODEL`	plan_exec.sh	Planner model; `qwen3.5:2b`
`EXEC_MODEL`	plan_exec.sh	Implementation model; `qwen2.5-coder:3b`
`PLAN_STYLE`	plan_exec.sh	`schema` by default; also `draft` or `twopass`
`SCHEMA_FILE`	plan_exec.sh	Project `scripts/handoff-schema.yaml`
`TMP_DIR`	devai.sh, plan_exec.sh	Generated code/plan location; `/tmp/devai`
`LOG_FILE`	devai.sh, plan_exec.sh	Project `logs/devai_history.log` or `logs/plan_exec_history.log`

ask.sh reads the deployment .env, then falls back to its own project-root .env; an exported API_KEY alone does not override that code path. The other two helpers use DEVAI_API_KEY first. All have an insecure default-key fallback, which should not be relied upon. Keep credentials out of command examples and shell history.

State	Location / behavior
ask.sh thread pointer	`/tmp/devsecops_conversation_${USER}.txt`
devai.sh thread pointer	`/tmp/devai_conversation_${USER}.txt`; `--clear` removes this pointer only
Planner output	`/tmp/devai/current_plan.txt`; overwritten by a new plan
Generated code	`gen_` or `exec_` files under `TMP_DIR`
Knowledge memory	Project `vector-store`, shared with backend RAG when roots match

plan_exec.sh does not persist a conversation ID in the way the other helpers do. It passes plan text between phases rather than isolating the YAML handoff field. If the plan exceeds 6,000 shell-counted characters, the implementation sends its first 6,000 bytes with a truncation notice; the complete plan remains on disk. Review whether important requirements were cut off, particularly with non-ASCII text. The complete request must still fit the API’s 20,000-character limit.

Flags and task text

Put flags before the task: bash scripts/plan_exec.sh --plan-only "task", not after it. The same applies to devai.sh --execute. Neither helper currently implements a general --model flag; use the variables above. ask.sh reads the question interactively.

Documentation commands

npm run docs:check
npm run build
npm run docs:links
npm run preview

The source check validates navigation, redirects, and page metadata. The built-link check inspects generated HTML and fragment targets. Neither command runs the backend or deploys the site.

Next: Use the CLI or contribute documentation.