OUR REPOSITORY. REAL EVIDENCE.

Clear instructions.
Three gaps worth closing.

Our repository already routes agents into a focused knowledge bundle. Inspection found a publishing-boundary issue, an unsafe operational instruction, and an untested assumption about agent discovery.

INSPECTED SNAPSHOT: First-party repo / 3b2678d
METHOD: File review + local image inspection
RESULT: 2 confirmed issues / 1 validation gap

MEDIUM PRIORITYCTX-01 / CONFIRMED

Internal knowledge ships with the public site.

Dockerfile:3        COPY . /usr/share/nginx/html
.dockerignore       No exclusion for AGENTS.md or knowledge/
Local image check   Both exist under the nginx document root

WHY IT MATTERS

Internal instructions and future operational notes become web-accessible. No secret or customer-data exposure was established.

RECOMMENDED CORRECTION

Copy only explicit public assets. Rebuild and verify that internal guidance returns 404.

This is a first-party demonstration on a small static website, not a customer engagement or representative benchmark. Findings describe snapshot 3b2678d, when this repository was named ContextOps. RepoChicken is the current service brand. CTX-01 and CTX-02 have been addressed in the redesigned prototype; CTX-03 remains unvalidated.

Download the Markdown report ↓Jump to the remediation plan →Repository context guide →

What we inspected

Root instructions, all six knowledge documents, site code, container configuration, and repository history. The running local container was confirmed to belong to this repository. Checks inspected file presence without running a coding agent against implementation tasks.

Snapshot inventory, measured before redesign
Surface	Measured baseline	Interpretation
Root AGENTS.md	33 lines / 321 words / 2,333 bytes	Instruction-file size, not measured token consumption
Knowledge bundle	6 Markdown files / 12,126 bytes	Available context, not proof it was loaded
Tracked repository	15 files	Small first-party demonstration scope
Agent environment	Not exercised	No cross-agent compatibility or behavioral result established

MEDIUM PRIORITYCTX-02 / CONFIRMED

Port troubleshooting grants too much authority.

knowledge/operations/local-preview.md:41
"If startup reports that port 8080 is already allocated,
stop the process currently using that port before retrying."

EVIDENCE & CONSEQUENCE

The playbook does not require identifying the port owner. An agent following it could stop an unrelated service. No interruption was observed or deliberately induced.

CORRECTION & ACCEPTANCE

Identify the owner first. Reuse this project's container, or select an explicit alternate port. Verify the playbook never directs an agent to stop an unrelated process.

Confidence: high in the instruction defect; downstream agent behavior was not tested.

VALIDATION GAPCTX-03 / NOT YET TESTED

The knowledge routes exist. Their use is unproven.

AGENTS.md:8        Routes readers to knowledge/index.md
knowledge/index.md Links the current architecture and operations concepts
Tracked snapshot  No recorded task trials, agent setup, or run outcomes

WHAT WE CAN CONCLUDE

The files provide an explicit discovery path. The reviewed snapshot contains no evidence that a particular agent consistently follows it. This does not establish that discovery fails.

HOW TO VALIDATE

Select an agent version and configuration. Repeat three representative tasks in fresh sessions on fixed snapshots. Record discovery, command selection, task checks, and failures before drawing a comparison.

Proposed trial tasks: change the email recipient; preview the site with an occupied port; update operational documentation after a simulated runtime change. Any commands must run in an agreed isolated environment.

A correction already visible in history

At 96446db, root AGENTS.md contained Compose commands. Commit 3b2678d replaced those with routing to the current playbook and added a maintenance trigger. This demonstrates a structural correction, not measured improvement in agent performance.

A sequenced remediation plan

Order	Action	Acceptance check	Prototype status
1 / CTX-01	Allowlist public image contents	Site and report return 200; AGENTS.md and knowledge/index.md return 404	Implemented
2 / CTX-02	Replace port-owner stop instruction	Owner identification and alternate-port path documented	Implemented
3 / CTX-03	Run controlled agent trials	Versioned setup, fixed tasks, repeated runs, recorded outcomes	Not performed

What this report does not claim

No customer code was reviewed. No token savings, productivity gain, task-success improvement, security assurance, or complete instruction-loading behavior was measured. The byte counts are filesystem measurements. This small repository does not establish delivery effort for a large monorepo.

The public audit proposal includes evidence review and recommendations. Remediation shown here is work on our own prototype, not an included customer implementation promise.

Could your team act on this?

That is the standard for the audit: a supported finding, a reason to care, and a correction with an acceptance check.

See the proposed pilot scope

Clear instructions.Three gaps worth closing.