The obvious thing that you are missing is that most of the world has no clue about how to do COSHH assessments properly, and so the substance-based approach that you mention is very common.  It's still wrong though and your approach is right.  Go for it!

My COSHH assessments are also tasked based as we sometimes require mixing substances, or use a substance in different ways such as one substance can be used to fill a machine, but can also be applied to the material using a brush, which are two different tasks and each brings their own risks.

What's even worse, I have seen companies provide MSDS's as a COSHH assessment.

As the same chemical used differently in two separate and different tasks can present very different hazards (as used)czxy only task based risk assessments stand any chance of being reliable indications of what is then needed to protect those who might be exposed.

Consider the case of that wonderful material called dihydrogen monoxide. Formed by combining two highly hazardous (flammable) gases but not then formally identified as hazardous. Long identified by dermatologists as an irritant to the skin but you will not usually find it on any safety data sheet. Actually, the cause of more occupational contact dermatitis than any other chemical. In certain conditions can cause severe skin burns or even be life threatening. In every workplace I have ever visited but often the risk of damage to skin is ignored.

http://www.dhmo.org/facts.html#USES

Some background reading on dihydrogen monoxide...

I'm in a similar position to knotty in that im at the point where im overhauling RA's for areas along with relevant COSHH assessments, so have the opportunity to start with a fresh process

There arent many (if any) situations were we use multiple substances simultaneously or combined in our processes, so i was thinking of adopting this approach:

1. Risk Assess the tasks
2. Identify substances within the tasks, and produce relevant COSHH assessment for the the exposure scenario (for example, disinfectant used in dilution, applied via mopping, exposure 1/2 an hour a day)
3. Within the RA, have reference to the scenario COSHH assessement. Actual cossh controls arent listed in the task RA.
4. SSW generated which goes into more detail on how to dilute, mop and safely conduct task incorporating the controls determined both in the RA and COSHH assessment.

Is there any flaws in this approach, or improvements that could be made?

"Ensure you refere to MSDS! - Yes but what about those chemicals that do not appear on the MSDS but are potentially hazardous, particularly when skin exposure is possible? For example the most authoritative list of skn sensitisers lists over 4900 chemicals, most of which will never have been classified as H317 and will therefor not appear on safety data sheers as such. Howeve, many of them will be found in workplaces. For example, when did you see dihydrogen monoxide (water) on a safety data sheet. Yet wet work is the most common cause of occupational contact dermatitis. It's the real hazard that arises when a task is carried out that must be the basis on which the risk assessment is carried out. Anything less runs a significant risk of an invalid risk assessment.

A lot has changed since the COSHH regulatiosn first got published and the whole approach needs to be reviewed. Why would you make an assessment of a chemical and leave our part of its reactive nature? Flammable hazards for example?

These days environmental issues are becoming critical, so again it should be part of the assessment process. Are the chemicals being used going to cause environmental impact? Containers may be recyclable, but how hazardous would be the process? Anyone who has experience of chemical disposal will tell horror stories of reactions occurring weeks after. I did clean up after a chemical fire, and we had to set a safety watch on the drums for about a month.

Task-based has to be the approach, but its how you integrate all the elements of the task.

Any chemist who has done process risk assessments will refer to how reactions change depending on the volumes of material and effects of pressure and temperature. A lab-scale process may be conducted safely but even scaled to 25 litres the reaction can be completely different.

For years I used to quote the example of tipex in training......the datasheet described the PPE required which is obviously appropriate for packaging on a production line but to ask an office staff member to wear, boots, gloves, and respiratory protection. 

Obviously, assessments change depending on how a material is used. 

From my experience assuming that because several people carry out the same task the risk for each of them will be the same and thus a generic risk assessment is a valid approach is somewhat doubtful.

Consider three people, each painting the same articles repeatedly. Now observe each. Quite often experience suggests that one will get almost no paint on them, one some and with the third it appears that the paint on the articles is subsidiary to the paint on them. Which level of exposure do you then base your risk assessment on?

We find the same situation when training workers to remove gloves without contaminating their hands with what is on the outside of the gloves. Even after having been shown how to do this correctly, some will be successful but commonly often over 50% fail the test revealing hands contaminated with the chemical. This is not unusual as this study illustrates…

“Forty-three hairdressers and apprentices were asked to remove gloves after washing hair. UV tracer was used to identify hand contamination before and after a training session on glove removal. All the participants (100%) had their hands contaminated during the first round.

In the second round 55.8% were contaminated. There were no significant differences between hairdressers and apprentices.” - Glove use among hairdressers: difficulties in the correct use of gloves among hairdressers and the effect of education, Oreskov KW, Søsted H, Johansen JD, Contact Dermatitis, 2015

With on-going training this can be reduced, but the reality is that you will never achieve identical results across a number of workers. So how does one allow for this in a risk assessment?

Isn’t this why our main effort should be in controlling the process rather than attempting to control the person?