agentleFS
r3dz4rr3dz4r/datapulse-my/docs/AGENTS.md

datapulse-my

AGENTS.md3 starsChanged 3 months agoMITView on GitHub

Working agreement for AI agents editing the public-facing documentation that auto-deploys to https://www.data-pulse.my. docs/ is the public website source. Cloudflare Pages publishes the assembled artifact from .github/workflows/deploy-cloudflare-pages.yml; every .html is served as-is. Most files are committed hand-authored source-of-truth documentation that drives the dashboard's content. Changes here are immediately visible to anyone visiting www.data-pulse.my. One-line constraint: the public-facing docs are the public-facing claim. Every word on data-pulse.my is part of the trust layer's contract with visitors.

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository6 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    No fabricated dataset IDs: When examples cite datasets like fuelprice, gtfs-static/prasarana?category=rapid-bus-kuantan, …

    line 15
  2. Distinctive:

    Tables: use real markdown tables, not bullets-with-em-dashes.

    line 41
  3. Distinctive:

    Numbers: always include units. a bare number is meaningless.

    line 43
  4. Very specific:

    When writing a codex brief that touches docs/:

    line 63
  5. Distinctive:

    Is the doc generated?: if yes, brief must say "regenerate via scripts/<gen>.py" not "edit the file"

    line 66
  6. Distinctive:

    Is the doc an audit/history file?: if yes, brief must create a NEW dated file, not edit existing

    line 67
  7. Distinctive:

    Does the change touch methodology_version?: STOP and surface to operator first

    line 68

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

AGENTS.md — docs/ (DataPulse MY public documentation)

Working agreement for AI agents editing the public-facing documentation that auto-deploys to https://www.data-pulse.my.

What this is

docs/ is the public website source. Cloudflare Pages publishes the assembled artifact from .github/workflows/deploy-cloudflare-pages.yml; every .html is served as-is. Most files are committed hand-authored source-of-truth documentation that drives the dashboard's content. Changes here are immediately visible to anyone visiting www.data-pulse.my.

One-line constraint: the public-facing docs are the public-facing claim. Every word on data-pulse.my is part of the trust layer's contract with visitors.

Hard rules

  1. Read-only-by-implication. Datapulse MY's docs describe a system that is itself read-only. Do not write prose that suggests writes, mutations, or upstream interactions beyond what the code actually does.
  2. Every external claim must be citeable. Numbers (the current dataset count, "10-status taxonomy") must be reproducible from datapulse.json + health/latest.json. The scheduler wakes every 5 minutes but probes only due datasets under tiered cadence; never describe this as all datasets every five minutes.
  3. No fabricated dataset IDs. When examples cite datasets like fuelprice, gtfs-static/prasarana?category=rapid-bus-kuantan, pharmaceutical_product_register, verify they exist in datapulse.json first. See scripts/check.py for the verifier.
  4. Methodology changes require an explicit version bump. health-methodology.md and health-methodology.html carry a methodology_version field (currently 3). Changes to the scoring formula, status taxonomy, or signal extraction require bumping this and updating consumers (mcp/server.py, dashboard rendering).
  5. The audit docs (AUDIT-*.md, DESIGN-AUDIT-*.md) are immutable history. They capture a point-in-time state. Add new audits, never edit old ones — even to fix typos.
  6. docs/index.html and docs/npra.html are GENERATED, not hand-edited. Edit scripts/embed_dashboard_data.py instead. The deterministic-safety-net gate will reject hand-edits that drift from the embedded data.
  7. Field notes (docs/field-notes/) are personal operator logs. Don't add formal structure (tables of contents, cross-references) — they're meant to be raw.
  8. No marketing superlatives. "Category-defining", "industry-leading", "first mover", "uncrowded" are all retired framings (see STATE.md 2026-08-17 Path A refresh). Use "intersection-defining" if needed, or stick to verifiable claims.

Doc taxonomy — which docs are hand-authored vs generated

FileTypeSourceRule
index.html, npra.htmlgeneratedscripts/embed_dashboard_data.pyNever hand-edit. The safety-net rejects drift.
health-methodology.htmlgeneratedscripts/gen_health_methodology.py from health/methodology.jsonNever hand-edit.
mcp-reference.mdgeneratedscripts/gen_mcp_reference.py from mcp/server.py ASTNever hand-edit.
release-verification.mdgeneratedscripts/verify_release_reproducible.pyCurrent proof only; includes source SHA, health freshness, dataset/tool counts, and protocol result.
architecture.md, branch-protection-handoff.md, mcp-deploy.md, operations.md, release-process.md, troubleshooting.mdhand-authoredoperator + Codex draftsStable until the operator explicitly revises.
AUDIT-*.md, DESIGN-AUDIT-*.md, health-compatibility-report-*.md, trust-snapshot-*.md, data-json-workspace-proposal-*.mdhand-authored, immutablepoint-in-time audit capturesImmutable. Add a new dated audit file; do not edit existing ones.
field-notes/*.mdoperator logoperator's running notesPersonal style; no enforced structure.
health-methodology.mdhand-authored source-of-truth for the generatoroperatorChanges here propagate to health-methodology.html via gen_health_methodology.py.
myaisafe-contract-inventory.md, contract-inventory.md, ai-directory-listings.md, adoption-seeding.md, record-evidence-v1.md, health-policy-compatibility.mdhand-authored, position papersoperator + CodexUpdate when the underlying claim changes. Cite sources.
trust-layer-notebook.ipynbhand-authored JupyteroperatorSee trust-layer-notebook.AGENTS.md.
trust-layer-notebook.AGENTS.mdhand-authored agent briefoperatorSee dedicated file.

Style conventions

  • Markdown files: sentence case headings. Bold for the one thing the section is teaching. No emoji in prose.
  • Tables: use real markdown tables, not bullets-with-em-dashes. Tables degrade to bullet groups when rich rendering is unavailable, but bullets don't degrade to tables.
  • Citations: link to https://data-pulse.my/... for self-references. Link to https://data.gov.my/... for upstream sources. Cite the date of the cited claim (as of 2026-08-17).
  • Numbers: always include units. a bare number is meaningless; a number with its unit is meaningful. "5-min" is ambiguous; "every 5 minutes" is unambiguous.
  • Code blocks: language-tagged (```bash, ```python, ```yaml). Don't use untyped fences.
  • Diagrams: ASCII for terminal-renderable contexts (Slack, Telegram, mobile browsers); SVG/Mermaid for the website itself, in assets/.

What is NOT in docs/

  • The dashboard JavaScript — that's bundled into __DATAPULSE_DATA__ inside index.html by embed_dashboard_data.py
  • Telemetry sinks — record-evidence/, data/<id>.json
  • The MCP server code — mcp/server.py
  • Pipeline artifacts — health/, deltas/

Out of scope

  • Adding new public-facing pages without operator approval — each public page is a new entry in the sitemap and a new inbound URL to defend
  • Editing generated docs by hand — the deterministic-safety-net gate will fail the next deploy
  • Removing historical audit docs — they're immutable history
  • Adding "trust us" framing without evidence — the operator's 2026-08-17 doctrine explicitly retires uncrowded/category-defining framings

Pre-flight checklist (for dispatch briefs)

When writing a codex brief that touches docs/:

  • Which file(s) change? path + line range
  • Is the doc generated? if yes, brief must say "regenerate via scripts/<gen>.py" not "edit the file"
  • Is the doc an audit/history file? if yes, brief must create a NEW dated file, not edit existing
  • Does the change touch methodology_version? STOP and surface to operator first
  • Are all numbers verifiable from the live data? brief must include the verification command (python3 scripts/check.py)
  • Does the change touch index.html or npra.html? brief must include python3 scripts/embed_dashboard_data.py
  • Workdir: absolute path to this repo

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.