capture-demand-signals
Skill4 starsChanged 3 months agoMITView on GitHub
Capture what customers are asking for into the workspace-global demand signal register, from a pursuit's requirement matrix or from conference / one-on-one / trial / debrief notes. Confirms a controlled capability theme and BD's status for each ask, then appends to signals/demand-signals.jsonl for the BD-to-CTO sync loop. Run after proposal-solution-architect on an active pursuit, and within 48 hours of any customer conversation.
In this repository30 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- adversarial-reviewSkill
- capture-intentSkill
- capture-portal-structureSkill
- capture-scorecardSkill
- capture-submissionSkill
- competitor-assessmentSkill
- compliance-checkSkill
- customer-intelSkill
- dashboardSkill
- evidence-checkSkill
- export-proposalSkill
- import-from-captureSkill
- narrative-spineSkill
- new-proposalSkill
- opportunity-quick-lookSkill
- past-performanceSkill
- pricing-analystSkill
- proposal-editorSkill
- proposal-graphicsSkill
- proposal-managerSkill
- proposal-patcherSkill
- proposal-solution-architectSkill
- proposal-storyboardSkill
- proposal-writerSkill
- red-team-reviewSkill
- setup-companySkill
- statusSkill
- submission-summarySkill
- technical-reviewSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Read docs/BD-CTO-DEMAND-SIGNAL-SYNC.md for the whole loop.
- Very specific: line 109
Quote, do not paraphrase: Put the customer's words in requirement.
- Very specific: line 153
evidence_idcloses the loop: When an ask is answered by something already in the ledger, cite it.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill 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.
Files
Capture Demand Signals
Purpose
Turn requirements the customer stated — in a solicitation, at a conference, across a table, during
a trial — into structured records in signals/demand-signals.jsonl, the workspace-global register
that feeds the CTO shop's view of demand.
This skill is the only writer of that register. It exists because the demand signal already lives in this workspace and never aggregates: forty pursuits hold their own requirement matrices, each with an honest Gap/Risk column, and none of them roll up. Engineering ends up hearing requirements as anecdotes from whoever was in the room last.
Read docs/BD-CTO-DEMAND-SIGNAL-SYNC.md for the whole
loop. This skill is the capture half; scripts/build-demand-signal-brief.py is the reporting half.
When to run
| Trigger | Mode | Timing |
| A pursuit's requirement matrix is written | --from-proposal <slug> | Right after /proposal-solution-architect, before drafting |
| Conference, industry day, one-on-one, trial, demo, or debrief | --from-notes <path> | Within 48 hours |
| A pursuit closes (win, loss, or no-bid) | --from-proposal <slug> | At close, to record bid_impact truthfully |
The 48-hour window on conversations is not bureaucratic. Verbal asks are the most valuable class in the register — they precede the written requirement by 6 to 18 months — and their value is in the customer's own words, which are gone from memory within a week.
Not this skill's job
- Tracking requirements inside one bid. That is
/compliance-checkand the compliance matrix. This register is cross-pursuit and exists for the roadmap conversation, not for the bid. - Recording submission mechanics. Page limits, cover pages, and POC blocks are not demand signals. They tell engineering nothing and dilute the channel.
- Deciding what engineering builds. BD records the ask and its cost to the bid. The disposition is engineering's, captured through the brief, never guessed here.
Inputs
reference/capability-themes.md— the controlled vocabulary. Required. Never invent a theme inline; if nothing fits, say so in the run summary and propose the addition as a separate edit.reference/schemas/demand-signal.schema.json— field contract.signals/demand-signals.jsonl— the existing register, to continuesignal_idnumbering.- Mode-specific source:
--from-proposal <slug>:proposals/<slug>/working/requirement-matrix.md(primary),working/quick-look.md,working/assumptions-and-risks.md,working/proposal-type.md--from-notes <path>: the note file, typically underproposals/<slug>/inputs/06_notes/orscrubs/
Procedure — --from-proposal <slug>
Step 1. Run the mechanical extractor.
python scripts/extract-demand-signals.py --proposal <slug>It parses the requirement matrix's tables, drops submission mechanics and internal pitch framing,
suggests a capability_theme by keyword, and infers our_status from the Gap/Risk prose. Output
lands in signals/candidates/<date>-candidates.jsonl.
Step 2. Confirm each candidate. The extractor is a first pass and says so: theme_suggestion
may be null, and our_status_inferred is unknown whenever the gap prose does not support an
inference. For each candidate, settle four things:
- Theme. Confirm the suggestion or replace it with a theme from the vocabulary. A null suggestion means classify it yourself. If the ask genuinely fits no theme, drop the candidate rather than forcing it.
our_status.shippingonly when a ledger item could prove it today.prototypewhen it exists but is not productized.roadmapwhen BD believes it is planned.gapwhen we could not answer.no-planfor a gap with no known intent to close it. Never leaveunknownin the register — resolve it or drop the row.demand_strength. Scored criterion, pass/fail, stated preference, or conversational.bid_impact. The most load-bearing field. What did this ask actually cost: prose hedged, criteria scored low, a teammate brought in, a question declined, a no-bid. Write it plainly and leave it null when the ask cost nothing. "Customer wanted X" is a wish; "we hedged 3.2 and teamed it to Scale" is a requirement.
Step 3. Merge duplicates against the existing register. The same customer asking the same thing
in a new solicitation is a new signal — recurrence is the point of the register. The same ask
appearing twice in one matrix is not. Check source_ref before appending.
Step 4. Append. One JSON object per line, signal_id continuing from the register's maximum,
never renumbering. Set schema_version to demand-signal.v1 and leave every cto_* field null —
those belong to engineering.
Step 5. Validate.
python scripts/build-demand-signal-brief.py --validate-onlyAn unknown theme is a hard failure: a typo silently splits a theme in two and understates its weight, which is the one error that corrupts the ranking the CTO shop is reading.
Procedure — --from-notes <path>
Same confirmation discipline, different extraction. There is no table to parse, so read the note and pull out only the sentences where the customer stated a need. Rules that matter here:
- Quote, do not paraphrase. Put the customer's words in
requirement. Verbatim language is what engineering trusts and what makes a recurring ask recognizable across three different agencies. date_observedis the date of the conversation, not today. The lead time between a verbal ask and its appearance in a solicitation is a number worth being able to compute later.- Attribute to the organization, not the individual.
customeris the agency or command; put the person innotesif it matters. demand_strength: curiosityis normal and valuable here. An ask nobody has funded yet is exactly the signal with the most roadmap lead time. Do not inflate it todesiredto make it look more important.proposal_slugmay be null. Relationship conversations often precede any pursuit.- Separate what the customer asked for from what we pitched. A note recording our own demo script is not a demand signal.
Output
Appended records in signals/demand-signals.jsonl. Report in the run summary:
- how many signals were appended, and the
signal_idrange - the theme breakdown
- how many carry
our_status: gaporno-plan(these become the CTO shop's queue) - any ask that fit no theme, with a proposed vocabulary addition
- anything intentionally dropped, and why
Activity trail
--from-proposal appends to that proposal's working/activity.md and working/ai-runs.jsonl per
the workspace contract. --from-notes with no proposal_slug has no proposal-level trail; note the
run in the register itself through each record's source_ref, which is the audit path that matters.
After capture
Nothing in this skill notifies anyone. The register is inert until someone builds the brief:
python scripts/build-demand-signal-brief.pyThe brief crosses each theme's demand against what my-company/evidence-ledger.json can
substantiate and leads with the five themes where that gap is widest. Two capture habits feed it
directly, and both are worth the extra seconds:
bid_impactdrives the ranking. Each hedged bid adds materially to a theme's exposure, and it is the field engineering reacts to. A theme with real demand and an emptybid_impactcolumn reads as academic.evidence_idcloses the loop. When an ask is answered by something already in the ledger, cite it. That is what separates "BD did not know we can do this" from "we genuinely cannot" — and the brief routes those two to different owners.
Distribute the generated .docx, never the .md — Drive does not take markdown, and the brief's
gap queue is the write-back surface the CTO shop fills in.
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
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.

