Person research mode
Person research mode turns a name into a scoped research run. It resolves the person’s identity, gathers public professional evidence from LinkedIn and the employer’s site, filters out namesakes, and adds a person profile to the normal research result. It is one specialized path within research skill execution and works with the planning, retrieval, quality, and output flows described in the related pages.
Person run activation
Person mode starts when a single-person topic uses --employer, --linkedin, or --email. The command requires a person topic and rejects comparison requests because identity lookup and evidence limits apply to one person at a time. The --company option has a different purpose: it researches a company and does not activate person mode. skills/…/last30days.py• L1019-1047 skills/…/last30days.py• L3404-3433
The run builds a person context. This context stores the name, identity anchors, lookup status, profile facts, employer bio, and later evidence counts. An employer name or domain and a work-email domain can anchor the run. A supplied LinkedIn URL is also an anchor. A personal-mail domain does not identify an employer. skills/…/person.py• L1-11 skills/…/person.py• L131-164
Resolution and evidence flow
The person resolve step gives LinkedIn and employer-site research one shared time limit. The two lanes run at the same time, and the main research flow continues with the results that finish before the limit. A slow or failed lane records a status instead of stopping the whole research run. skills/…/last30days.py• L3436-3487 skills/…/last30days.py• L3489-3519
The diagram shows how person inputs become verified identity context and then feed the normal evidence pipeline.
flowchart LR
A[Person topic and identity input] --> B[Person context]
B --> C[LinkedIn resolution]
B --> D[Employer site bio]
C --> E[Identity anchors and profile facts]
D --> E
E --> F[Planner and source queries]
F --> G[Identity filter and ranking]
G --> H[Person profile and findings]
Planner activation
Person context changes the research plan in two ways. Its anchors join any anchors from the plan, and the plan is marked as a person run. The identity gate then remains closed when no reliable anchor exists, so a shared name alone cannot make a result pass as evidence about the person. This joins person-specific identity work to the general planning and query flow without creating a separate report pipeline. skills/…/pipeline.py• L2593-2600 skills/…/planner.py• L242-260
For externally supplied plans, person_identity_anchors carries explicit identity terms. The planner validates these as non-empty strings and preserves them through follow-up plans. This lets a host provide a known organization or other identity clue while avoiding invented affiliations. skills/…/planner.py• L257-260 skills/…/planner.py• L425-433
References: skills/…/pipeline.py• L2593-2600 skills/…/planner.py• L242-260 skills/…/planner.py• L425-433
LinkedIn resolution
LinkedIn resolution prefers a profile supplied by the user. When no profile is supplied, an enabled lookup can propose a profile from the person’s name and employer, or use the full email only when no employer is available. Candidate profiles must show a current or recent employer match before the engine accepts them, except for a profile found directly through email lookup. A rejected namesake does not add an anchor. skills/…/person_linkedin.py• L1-24 skills/…/person_linkedin.py• L227-259 skills/…/person_linkedin.py• L262-329
The lane records both the profile facts and the reason for the match. Its status distinguishes a checked profile from a missing key, provider failure, missing profile, or rejected match. A current employer that differs from the supplied employer is recorded as a mismatch rather than silently treated as the same employment record. skills/…/person_linkedin.py• L140-191 skills/…/person_linkedin.py• L333-360
References: skills/…/person_linkedin.py• L1-24 skills/…/person_linkedin.py• L140-191 skills/…/person_linkedin.py• L227-259 skills/…/person_linkedin.py• L262-329
Employer-site evidence
The employer-site lane finds a public employer host from an explicit domain, a non-personal email domain, or an employer-name lookup. It checks common public people pages and likely person-page paths. An optional site-scoped search can add public pages after those fixed paths do not find the person. skills/…/employer_site.py• L1-17 skills/…/employer_site.py• L298-345 skills/…/employer_site.py• L383-405
An employer page counts only when it contains the person’s full name. The lane returns the nearby title and bio text, plus links from the person’s own page block. It does not treat a company header, footer, or a colleague’s card as the person’s evidence. skills/…/employer_site.py• L149-193 skills/…/employer_site.py• L259-282
References: skills/…/employer_site.py• L1-17 skills/…/employer_site.py• L149-193 skills/…/employer_site.py• L259-282 skills/…/employer_site.py• L298-345 skills/…/employer_site.py• L383-405
Identity boundaries
Identity anchors control which public items can enter person findings. The final pipeline filters each source after retrieval, and the ranking stage applies the same identity rule before scoring. A source item must name an anchor or carry a trusted first-party marker. Only the explicitly resolved X subject can pass without a text anchor; a related or discovered account cannot inherit that trust. skills/…/pipeline.py• L3404-3428 skills/…/rerank.py• L287-297 skills/…/entity_extract.py• L167-210
Links found in a verified LinkedIn profile or employer bio can fill missing X and GitHub handle inputs. The resolver skips links that look like employer accounts, social sites, or employer domains. These discovered handles support first-party evidence, but they do not replace the identity checks for other source material. skills/…/person.py• L266-304 skills/…/person.py• L307-325
Privacy boundaries
The email address is a lookup input only. The run never uses the full address as an identity anchor, log value, or serialized field. Only a non-personal email domain can support employer identity. skills/…/person.py• L120-128 skills/…/person.py• L177-200
Employer evidence stays on a public employer host. Fetches reject private hosts and off-site redirects. Remote page text is shortened and marked as untrusted before it reaches the host, which keeps public bio content separate from control data. skills/…/employer_site.py• L238-247 skills/…/employer_site.py• L348-368 skills/…/employer_site.py• L407-416
References: skills/…/person.py• L120-128 skills/…/person.py• L177-200 skills/…/person.py• L266-325 skills/…/employer_site.py• L238-247 skills/…/employer_site.py• L348-368
Person-profile output
The pipeline stores the resolved person context as a person_profile artifact after identity filtering. The artifact joins identity basis, LinkedIn facts, employer bio data, discovered links, activity counts, dropped namesakes, and lane status with the ranked research findings. skills/…/pipeline.py• L3407-3428 skills/…/person.py• L177-200
The renderer places a Person Profile section before ranked evidence when the artifact exists. It shows whether identity is certain, current roles and headline, the employer bio, links found on profile pages, LinkedIn status, activity coverage, older activity when the current window is thin, and counts of items dropped as namesakes. Unchecked or timed-out work stays visible as unknown or incomplete rather than appearing as a clean zero. skills/…/render.py• L2590-2633 skills/…/render.py• L2635-2673 skills/…/render.py• L2675-2729
References: skills/…/pipeline.py• L3407-3428 skills/…/person.py• L177-200 skills/…/render.py• L2590-2633 skills/…/render.py• L2635-2729
Change checklist
When these files or sections are changed, remember to consider these:
- Keep person input validation aligned with the context fields that become anchors and the privacy rules for email values.
- Keep LinkedIn match reasons and statuses aligned with the profile data shown in the person-profile block.
- Keep employer-site host checks aligned with the promise that only public employer pages and safe redirects are read.
- Keep identity filtering aligned in both final source gating and ranking so namesakes cannot return through a high score.
- Keep discovered X and GitHub handles marked as discovered links rather than treating them as user-supplied subject identity.
- Keep timeout and unknown states distinct from zero activity in the rendered profile.
Important failure modes
| Trigger | Consequence | Guard |
|---|---|---|
| Person flags are used without a person topic or with a comparison request | The command stops before lookup and paid calls | Single-person input gate |
| A shared name has no employer, work domain, or supplied LinkedIn anchor | Namesake evidence is rejected and identity is shown as uncertain | Person-run identity gate |
| LinkedIn lookup finds a profile whose current or recent role does not match | The candidate is rejected instead of becoming an anchor | Employer and domain match check |
| Employer page redirects to a private or different host | The page is not used as person evidence | Public-host and redirect checks |
| A resolve lane reaches the shared deadline | The report continues with a timeout or unknown status | Shared deadline and explicit status rendering |
| Remote bio text contains control-like content | The text is treated as data and limited before rendering | Untrusted-content fence and output caps |
Key files
| File | Why to read it |
|---|---|
| skills/…/person.py | Defines person context, identity anchors, email privacy, and discovered handles. |
| skills/…/person_linkedin.py | Defines LinkedIn lookup, profile verification, match status, and mismatch reporting. |
| skills/…/employer_site.py | Defines public employer-site discovery, page evidence, and host safety limits. |
| skills/…/last30days.py | Shows person-mode activation and the concurrent resolve step. |
| skills/…/pipeline.py | Shows planner activation, identity filtering, and person-profile artifact creation. |
| skills/…/entity_extract.py | Defines the fail-closed identity test for source items. |
| skills/…/render.py | Defines the rendered person profile and its uncertainty signals. |