name	agent-adrs
description	Consult the agent-adrs catalog of engineering decision records before writing code involving time, money, numerics, text encoding, identifiers, validation, errors, observability, concurrency, data storage, caching, networking, APIs, security, configuration, dependencies, testing, files, or i18n. Apply rules per their consensus tier and cite them by id.

agent-adrs: applying shared decision records

This project follows agent-adrs: generalized, evidence-backed engineering rules in ADR format.

When to consult

Before writing or reviewing code that touches any covered category (see the description above). Signals include: datetime construction, currency amounts, float comparisons, string encoding/length, ID generation, input parsing, error handling, log statements, shared state, schema/migration changes, cache logic, outbound network calls, API handlers, secrets, config loading, dependency additions, test code, file writes, locale-dependent formatting.

How to consult (two layers)

Index first (cheap): read index/INDEX.md (or just the relevant index/<category>.md) from the local clone/submodule, or from https://raw.githubusercontent.com/gmoigneu/agent-adrs/main/index/INDEX.md. Each line is id [consensus-tier] rule.
Full ADR on demand: when a rule is relevant to the change at hand, fetch adrs/<category>/<NNNN>-<slug>.md for the Context, Decision, boundaries, Enforcement pointers, and Exceptions before applying it.

Skip rules whose status is deprecated or superseded. Follow the superseded_by pointer instead.

Consensus tiers: behavior contract

established: apply by default. Deviate only on explicit project instruction, and say that you are deviating and why.
recommended: apply unless the project signals otherwise (an existing contrary convention, configuration, or instruction). When not applying, note the deviation in your response.
contested: NEVER auto-apply. Surface the trade-off from the ADR's Context and ask the user, or follow the project's existing choice.

Citing and excepting

Cite applied rules by id in explanations and review comments: agent-adrs/time-0001.
When code must legitimately violate a rule, add the inline suppression comment so the exception is visible and greppable: adr-exception: <id> -- <reason>
When reviewing diffs, flag violations of established and recommended rules that lack a suppression comment.

agent-adrs: applying shared decision records

This project follows agent-adrs: generalized, evidence-backed engineering rules in ADR format.

When to consult

How to consult (two layers)

Index first (cheap): read index/INDEX.md (or just the relevant index/<category>.md) from the local clone/submodule, or from https://raw.githubusercontent.com/gmoigneu/agent-adrs/main/index/INDEX.md. Each line is id [consensus-tier] rule.

Full ADR on demand: when a rule is relevant to the change at hand, fetch adrs/<category>/<NNNN>-<slug>.md for the Context, Decision, boundaries, Enforcement pointers, and Exceptions before applying it.

Skip rules whose status is deprecated or superseded. Follow the superseded_by pointer instead.

Consensus tiers: behavior contract

established: apply by default. Deviate only on explicit project instruction, and say that you are deviating and why.

recommended: apply unless the project signals otherwise (an existing contrary convention, configuration, or instruction). When not applying, note the deviation in your response.

contested: NEVER auto-apply. Surface the trade-off from the ADR's Context and ask the user, or follow the project's existing choice.

Citing and excepting

Cite applied rules by id in explanations and review comments: agent-adrs/time-0001.

When code must legitimately violate a rule, add the inline suppression comment so the exception is visible and greppable: adr-exception: <id> -- <reason>

When reviewing diffs, flag violations of established and recommended rules that lack a suppression comment.