What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A configuration file can exist in your repository without being loaded by the coding agent that handled your task. To find the cause, reproduce the issue in the same agent and interface, verify the repository and file path, then look for evidence that the agent loaded the instructions. Treat loading and following as separate questions: a file reference proves it entered context, not that every rule was obeyed.
Why is my AI coding agent ignoring my instructions?
Discovery depends on the agent, the surface you use (such as an IDE, CLI, or chat), the file’s exact name and location, and sometimes the files involved in the task. A rule may also be absent from the session or may conflict with a more local or higher-priority instruction. Start by reproducing the failure in the same context; checking another checkout or a different interface can send you looking for the wrong cause.
Run the 15-minute audit
The time bands below are suggested allocations, not a measured completion time. Use them to keep the investigation focused; some cases will take longer.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Minutes 0–2: Reproduce the failure precisely
- Record the agent and product surface: for example, the specific IDE, CLI, cloud task, or chat where the behavior occurred.
- Confirm the repository or workspace and its root, the current directory, and the target file path.
- Choose a small task that reliably shows the unexpected behavior. Keep the reproduction in the same repository and surface as the original failure.
Codex recommends checking its reported workspace root and asking it to list active instruction sources. See OpenAI’s Codex AGENTS.md documentation.
#1 Best Overall
Minutes 2–5: Confirm the expected file
- Check the exact filename, spelling, capitalization, location, and that the file contains the rule you expect.
- Verify that this agent recognizes that filename and format on the surface you are using. Do not assume a filename supported in one mode is supported in another.
- If you renamed the file or chose a fallback filename, check whether the agent is configured to discover it. In Codex, fallback filenames are ignored unless configured.
Minutes 5–9: Check scope, precedence, and limits
Inspect the repository root and relevant ancestor and nested directories. Look for additional instructions that could change which rule applies, including overrides, path-specific rules, and user- or organization-level instructions. A file can be valid but out of scope for the task, or superseded by instructions closer to the files being edited.
Codex: inspect its project instruction chain
OpenAI documents Codex as combining project instruction files from the root down, with nearer files appearing later in the combined prompt. Check for nested AGENTS.md and AGENTS.override.md files, as well as user-level instructions. The documented default combined project instruction limit is 32 KiB, so guidance beyond that limit may be omitted. Codex also supports project_doc_max_bytes and configurable fallback filenames. If you change Codex configuration, restart it or run it again so the change is loaded. The details are in OpenAI’s Codex AGENTS.md documentation.
GitHub Copilot: match the instruction type to the task
GitHub documents distinct instruction files and scopes:
.github/copilot-instructions.mdprovides repository-wide instructions..github/instructions/NAME.instructions.mdprovides path-specific instructions usingapplyToglobs. Confirm that the pattern matches the files involved in the task.AGENTS.mdprovides agent instructions; the nearest file takes precedence.
Matching repository-wide and path-specific instructions may both apply. Support varies by Copilot feature or surface, so confirm the relevant behavior for the one you used. See GitHub’s repository instructions documentation.
Other agents: verify their current rules
Do not assume Codex or Copilot’s filenames, scope, precedence, glob behavior, or verification methods apply to another product. Check that agent’s current documentation for the exact product and surface. Discovery rules can change, so older guidance may not describe the current behavior.
Minutes 9–12: Look for observable evidence of loading
Use the agent’s own documented way to inspect active instructions. In Copilot Chat, check the response’s reference list for the repository instruction file. In Codex, ask it to list active instruction sources or use the documented TUI log approach. A missing reference or source is a reason to investigate discovery and scope; a visible one is evidence of loading, not proof of adherence.
Rank #4
Minutes 12–15: Run a discriminating test
Add or identify one harmless, unique instruction whose effect is easy to recognize, then ask a narrowly framed question or task that depends on it. Compare the result with the file and the agent’s source evidence. Do not use secrets or risky commands as diagnostic instructions.
- If there is no evidence the file loaded: return to the exact filename, root, scope, matching path, overrides, and any product-specific limits or configuration.
- If the file loaded but the rule was not followed: inspect whether the rule is clear, relevant to the task, and in conflict with another instruction. Try a clearer, actionable version as a separate test.
Does loading a config file mean the agent will follow it?
No. Loading and adherence are different diagnostic questions. A source reference or active-instruction report can show that instructions entered the context; it cannot establish that every instruction was followed. If the source is present, focus next on clarity, conflicts, relevance, and whether the rule describes an actionable behavior. Evaluate the specific behavior with a reproducible test rather than treating file discovery as a guarantee.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
