SEBI Compliance Research

Source-first research skill for mapping SEBI, NSE/BSE, and selected Companies Act obligations of Indian listed companies. Method, coverage controls, and evals only — no registers, no client data.

This is / this is not

This is

• A reusable six-phase AI research skill (SKILL.md) plus reference files and evaluation cases.
• A method for building a traceable 12-column compliance register from live primary sources.
• A coverage-and-quality control pack: 19-category MECE checklist, known-omission tripwires, seven pre-delivery gates.

This is not

• Legal, regulatory, tax, or investment advice.
• A SEBI circular archive, scraper, monitor, or dataset.
• A filled-in Excel register you can download and file from.
• A substitute for a Company Secretary, compliance officer, or qualified counsel.

Stored URLs in this repository are discovery seeds. They are not proof that a rule is current.

Why this exists (the problem it solves)

Ad-hoc web research and unconstrained AI runs on Indian listed-company compliance fail in predictable ways:

1. They cite a stale consolidated regulation instead of the version on SEBI’s listing page today.
2. They treat law-firm blogs as primary sources.
3. They build coverage from whatever search returns, not from a checklist constructed before searching.
4. They read snippets instead of full regulations and schedules.
5. They skip the last-12-month amendment sweep.
6. They treat NSE/BSE compliance calendars as secondary. Those calendars contain listing-condition items that appear in no SEBI document.
7. They have no domain omission tripwires, so the same filings are missed every time.

An early register produced without these controls missed 18 filings and contained 5 errors. This repository exists so the next run is planned, source-first, MECE-complete, exchange-reconciled, and gated before delivery.

Who it is for

• Company Secretaries and listed-company compliance teams who need a filing register or calendar that can be audited to primary URLs.
• Compliance heads and GRC advisors who must map LODR / PIT / SAST / ICDR / NCS / buyback / delisting / SBEB plus exchange calendars.
• AI-agent builders who want a skill file they can load, evaluate, and run against live web retrieval — not a hallucinated obligation list.

Not for: extracting obligations from an uploaded regulatory PDF as the sole source of truth, or analysing a court or tribunal judgment. Those are different workflows. This skill may still supply the MECE checklist and omission tripwires as a completeness audit.

Required capabilities

To run the research (not merely read the method) the host environment must provide:

• Live web search.
• Full-page retrieval of official pages and linked PDFs on sebi.gov.in, nseindia.com, bseindia.com, and mca.gov.in.

Optional: a crawler that can extract JavaScript-rendered listing pages. If a primary page cannot be fetched, the skill records the row as UNVERIFIED. It must not silently substitute a blog.

Validating the repository files only needs Python 3. It does not run the research.

Who should do what

You want to…	Do this
Understand the method without an agent	Read this README, then SKILL.md Parts 0–2 and Phases 1–6, then the four files in references/ in the order given below. Stop there. Do not treat seed URLs as current law.
Run the research	Install or load the skill (next section), confirm live web search + page fetch are available, then use the prompt in Run the research.
Check that the public tree is intact	python3 scripts/validate_repo.py

Install / load the skill

Clone the public tree:

git clone https://github.com/d33pm3/SEBI-Compliance-Research.git
cd SEBI-Compliance-Research

Keep this layout together. SKILL.md is the entry point. Phase files live under references/. Do not load SKILL.md alone and expect the checklists to be in context.

Copy the skill into a skill-compatible host (create the folder if it does not exist):

sebi-india-compliance-research/
├── SKILL.md
├── references/
│   ├── canonical_sources.md
│   ├── known_omissions.md
│   ├── mece_checklist.md
│   └── output_standards.md
└── evals/
    ├── evals.json
    └── trigger-evals.json

Typical drop paths (names vary by product version; use the host’s current skills directory):

• Claude Skills / Claude Code: project or user skills/sebi-india-compliance-research/
• Cursor: .cursor/skills/sebi-india-compliance-research/ or the project skills folder
• Grok custom skills: the configured custom-skills directory, or paste/load SKILL.md and open the matching references/ file when the phase says to
• Other skill hosts: any folder the product indexes as a skill, with SKILL.md at the root of that folder

Then open a new agent session so the host re-indexes skills.

Viewing on GitHub does not load the skill into an agent. Cloning or copying the files does not grant a copyright license to reuse the materials in a product. See View vs use.

Run the research

1. Confirm the session can search the web and fetch full pages.
2. Load this skill (SKILL.md).
3. Fix entity scope before the first fetch: equity-listed / debt-listed / HVDLE / SME / market-cap tier, and the deliverable (register, calendar, gap analysis, or memo).
4. Paste a prompt like the following.

Use the sebi-india-compliance-research skill.

Entity: NSE + BSE equity-listed company, Top 500 by market cap.
Deliverable: 12-column SEBI / NSE / BSE compliance register in Markdown, then XLSX if the environment can write files.

Execute Phases 1–6 in order.
- Phase 1: populate the Canonical Source Table from live listing pages. Seed URLs are search targets only.
- Phase 2: Tier 1 regulator sources, then independent NSE and BSE calendar passes.
- Phase 3: map all 19 MECE categories before broad search.
- Phase 4: audit against known-omission tripwires.
- Phase 5: every row uses the locked 12-column schema.
- Phase 6: run all seven gates. Label each legal value FETCHED or UNVERIFIED.

Do not cite law-firm blogs as Source URL. Do not invent timelines. If a primary page cannot be fetched, mark the row UNVERIFIED and list the URL I should open.

5. When the skill names a phase reference, open only that file (references/canonical_sources.md, mece_checklist.md, known_omissions.md, or output_standards.md).
6. Treat the output as a research draft. Re-fetch primary sources before operational use. Have a CS or counsel review it.

Repository structure checks (not research):

python3 scripts/validate_repo.py

Sample 12-column output (illustrative only)

The table below shows the shape of the deliverable. Values are not current law. Do not copy them into a live register. Every operational row must be re-read from a primary URL fetched in that session.

S.No.	Category	Filing / Disclosure Name	Regulation Reference	Applicable To	Filing Authority	Frequency	Trigger / Event	Timeline from Trigger	Format / Mode	Penalty for Non-Compliance	Source URL
1	Financial results	Quarterly financial results	LODR Reg 33(3)(a)	Equity-listed	NSE / BSE	Quarterly	End of each quarter	Outer bound from the fetched consolidated text	XBRL + PDF as specified by the exchange	Exact provision from the fetched text	Live sebi.gov.in URL fetched this session
2	Board / committee	Outcome of board meeting	LODR Reg 30 read with Schedule III	Equity-listed	NSE / BSE	Event-driven	Board decides specified matters	Dual branch: during vs after market hours, as fetched	Exchange portal + PDF	Exact provision from the fetched text	Live sebi.gov.in URL fetched this session
3	Insider trading	Initial disclosure by designated person	PIT Reg 7(1)	Listed company + designated persons	Company / stock exchanges as fetched	One-time / event-driven	Becoming a designated person / joining	Trading-day outer bound from the fetched PIT text (do not use blog timelines)	Prescribed form as fetched	Exact provision from the fetched text	Live sebi.gov.in URL fetched this session
4	Exchange calendar item	Item that appears on the NSE or BSE compliance calendar	Listing condition / circular cited on the calendar	As stated on that calendar row	NSE or BSE	As per calendar	As per calendar	As per calendar, then reconciled to the SEBI instrument	As per calendar	As fetched	Live nseindia.com or bseindia.com URL fetched this session

Verification labels used in a real run: FETCHED (read from a primary page this session) or UNVERIFIED (fetch failed or not yet read). Skill-file text is never a citable legal source.

What the method does

• Verifies current consolidated regulations and recent amendments before extraction.
• Builds coverage from a 19-category MECE checklist rather than search-result recall.
• Reconciles SEBI-derived obligations against NSE and BSE compliance calendars.
• Produces traceable 12-column compliance registers with explicit verification status.
• Applies known-omission tripwires and seven pre-delivery quality gates.

Repository structure

.
├── .github/workflows/validate.yml  # Structural and safety checks
├── evals/
│   ├── evals.json                  # Output-behaviour evaluations
│   └── trigger-evals.json          # Invocation-routing evaluations
├── references/
│   ├── canonical_sources.md        # Primary-source discovery starting points
│   ├── known_omissions.md          # Historical omission tripwires
│   ├── mece_checklist.md           # Coverage model
│   └── output_standards.md         # Register and delivery requirements
├── scripts/validate_repo.py        # Local validation entry point
├── .gitignore                      # Public-repository exclusion controls
├── LICENSE                         # Viewable-source proprietary terms
├── SECURITY.md                     # Vulnerability and data-exposure reporting
└── SKILL.md                        # Skill entry point and six-phase workflow

Reading order if you are not running an agent:

1. This README
2. SKILL.md (Parts 0–2, then Phases 1–6)
3. At Phase 1: references/canonical_sources.md
4. At Phase 3: references/mece_checklist.md
5. At Phase 4: references/known_omissions.md
6. Before delivery: references/output_standards.md

Scope and limitations

This project supports regulatory research. It does not provide legal advice and does not replace review by qualified legal or compliance professionals. Regulations, circulars, exchange requirements, and enforcement positions change. Any operational use must re-fetch and validate primary sources during the current research session.

The repository contains the research method, coverage controls, source-discovery starting points, and evaluation cases. It intentionally contains no regulatory source documents, client data, credentials, vector stores, embeddings, generated registers, or internal reports.

Security and privacy

Do not commit confidential documents, customer or client information, audit findings, credentials, private prompts, RAG corpora, vector-store exports, embeddings, or production configuration. See SECURITY.md and .gitignore.

View vs use (plain-language license)

Viewing is allowed. Using is not a granted right.

• You may open this public repository and read the files.
• Public visibility is not an open-source license.
• You may not copy, modify, distribute, sublicense, sell, or commercially use the contents unless DK Mendiratta has given prior written permission, or applicable law already allows that act.
• Materials are provided “as is”, without warranties. You remain responsible for verifying current primary sources and obtaining professional review.

The governing text is in LICENSE.

GitHub About box (sidebar)

Use these values on the repository About form so search cards match this README:

• Description: Source-first research skill for mapping SEBI, NSE/BSE and selected Companies Act obligations of Indian listed companies. Method, coverage controls and evals only — no registers, no client data.
• Topics: sebi lodr compliance india grc nse bse listed-companies ai-skills regulatory-mapping
• Website: leave empty unless a project site is published.
• License: proprietary (see LICENSE; GitHub will show “Other”).