Reads before it writes
read_file and grep open your scripts and logs, so a new script follows the habits already in your folder.
Product
Bash language work, from first draft to reviewed diff
01
You bring the Bash language, and TechRetriever brings the tools: it reads your scripts, drafts new ones, runs them in WSL or over SSH on systems you own, and shows the real result of each step as a card. Plan only runs nothing. Version 2.9.30 needs Windows 10 or 11.
Runs and reports
Shell runs your script and reports the real exit code, so a failed command shows up as a failed card.
You approve the change
edit_file shows a diff, and Ask first pauses before actions that change something.
What the agent does with Bash scripts
- Filesread_file, write_file, edit_file, list_dir, glob and grep work inside the chat's working folder, and a counter shows files created, edited and read.
- Shellruns commands such as bash -n and your own test command, and judges success by the real exit code. Dangerous commands are blocked.
- Webweb_search finds pages and http_get reads official manual pages, so an explanation can point to its source. Search can run over an anonymizing network.
- Screenshots and imagesocr reads the text of an error message from a screenshot, and gen_image saves image files.
- Orchestrationtask starts parallel sub-agents, run_on_model sends one hard step to a bigger model, and skill keeps reusable playbooks.
- Approval modesAuto-run acts at once, Ask first pauses before actions that change something, and Plan only writes a plan and runs nothing.
- Cronruns a prompt on a schedule, for example a recurring review of a log folder. Plans survive restarts.
- Modelsconnect a cloud AI provider, a model running locally on your PC or the built-in model, then press Test to see response time and context window.
Where Bash fits and where the agent helps
Treat the Bash programming language as glue that joins commands and files for automation, not for heavy calculation. Keep scripts short. The agent drafts the glue, and you decide which jobs deserve one.
Count the .log files in my logs folder, explain what my count script does, and add a usage message.
- glob
logs/*.log4 files - read_file
scripts/count-logs.sh12 lines - grep
usage scripts0 matches - edit_file
scripts/count-logs.shdiff shown - Shell
bash -n scripts/count-logs.shexit 0 - Shell
bash scripts/count-logs.sh logsexit 0
Waiting for your OK before the next stepRejectAccept
Two jobs, start to finish
Add a usage message to an existing script
Ask the agent to add a usage message to count-logs.sh so the script stops when no folder is given. read_file loads the script, and grep checks whether a usage line exists already. Beginners can ask the agent to explain the Bash commands in count-logs.sh that read the file system before they edit anything. edit_file then adds a short test and shows the change as a diff. Because set -u stops the script when $1 is unset, the test should check $# first. Read the diff, accept it only if that check comes before any use of $1, and run bash -n on the result.
Add shortcuts to your own shell
A simple directory jumper for Bash is a short function that changes to a folder you saved under a name. Ask the agent to write it with write_file, and let web_search and http_get read the official manual page for the function and alias syntax it uses. Keep it small. Learning Bash is largely learning scripting patterns, and a jumper that walks through a list of saved names is a natural place to meet while loops. Test the function in a throwaway shell inside WSL rather than in your main session. Read the diff of every change to your startup file, and keep a backup copy of that file before you accept anything.
What TechRetriever does not do
TechRetriever is not a hosted service or a plugin. It is meant only for your own files, your own PC and servers you are authorised to use, and it does not enforce that scope, so you keep each job inside it. Nothing here guarantees that code, scans or analysis are correct.
